fix: use Alpine-compatible addgroup/adduser flags in Dockerfiles

Replace --system/--gid/--uid (Debian syntax) with -S/-g/-u (BusyBox/Alpine). Coolify ARG injection causes exit code 255 with Debian-style flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
refactor(coolify): externalize postgres, qdrant, S3
2026-03-06 22:38:31 +01:00 · 2026-03-03 09:23:32 +01:00 · 2026-02-25 10:43:15 +01:00 · 2026-02-15 18:36:38 +01:00 · 2026-02-15 17:32:24 +01:00 · 2026-02-15 16:58:30 +01:00
113 changed files with 24536 additions and 6038 deletions
--- a/.env.coolify.example
+++ b/.env.coolify.example
@@ -0,0 +1,79 @@
+# =========================================================
+# BreakPilot Lehrer — Coolify Environment Variables
+# =========================================================
+# Copy these into Coolify's environment variable UI
+# for the breakpilot-lehrer Docker Compose resource.
+# =========================================================
+
+# --- External PostgreSQL (Coolify-managed, same as Core) ---
+POSTGRES_HOST=<coolify-postgres-hostname>
+POSTGRES_PORT=5432
+POSTGRES_USER=breakpilot
+POSTGRES_PASSWORD=CHANGE_ME_SAME_AS_CORE
+POSTGRES_DB=breakpilot_db
+
+# --- Security ---
+JWT_SECRET=CHANGE_ME_SAME_AS_CORE
+
+# --- External S3 Storage (same as Core) ---
+S3_ENDPOINT=<s3-endpoint-host:port>
+S3_ACCESS_KEY=CHANGE_ME_SAME_AS_CORE
+S3_SECRET_KEY=CHANGE_ME_SAME_AS_CORE
+S3_BUCKET=breakpilot-rag
+S3_SECURE=true
+
+# --- External Qdrant (Coolify-managed, same as Core) ---
+QDRANT_URL=http://<coolify-qdrant-hostname>:6333
+
+# --- Session ---
+SESSION_TTL_HOURS=24
+
+# --- SMTP (Real mail server) ---
+SMTP_HOST=smtp.example.com
+SMTP_PORT=587
+SMTP_USERNAME=noreply@breakpilot.ai
+SMTP_PASSWORD=CHANGE_ME_SMTP_PASSWORD
+SMTP_FROM_NAME=BreakPilot
+SMTP_FROM_ADDR=noreply@breakpilot.ai
+
+# --- LLM / Ollama (optional) ---
+OLLAMA_BASE_URL=
+OLLAMA_URL=
+OLLAMA_ENABLED=false
+OLLAMA_DEFAULT_MODEL=
+OLLAMA_VISION_MODEL=
+OLLAMA_CORRECTION_MODEL=
+OLLAMA_TIMEOUT=120
+
+# --- Anthropic (optional) ---
+ANTHROPIC_API_KEY=
+
+# --- vast.ai GPU (optional) ---
+VAST_API_KEY=
+VAST_INSTANCE_ID=
+
+# --- Game Settings ---
+GAME_USE_DATABASE=true
+GAME_REQUIRE_AUTH=true
+GAME_REQUIRE_BILLING=true
+GAME_LLM_MODEL=
+
+# --- Frontend URLs (build args) ---
+NEXT_PUBLIC_API_URL=https://api-lehrer.breakpilot.ai
+NEXT_PUBLIC_KLAUSUR_SERVICE_URL=https://klausur.breakpilot.ai
+NEXT_PUBLIC_VOICE_SERVICE_URL=wss://voice.breakpilot.ai
+NEXT_PUBLIC_BILLING_API_URL=https://api-core.breakpilot.ai
+NEXT_PUBLIC_APP_URL=https://app.breakpilot.ai
+NEXT_PUBLIC_STRIPE_PUBLISHABLE_KEY=
+
+# --- Edu Search ---
+EDU_SEARCH_URL=
+EDU_SEARCH_API_KEY=
+OPENSEARCH_PASSWORD=CHANGE_ME_OPENSEARCH_PASSWORD
+
+# --- Misc ---
+CONTROL_API_KEY=
+ALERTS_AGENT_ENABLED=false
+PADDLEOCR_SERVICE_URL=
+TROCR_SERVICE_URL=
+CAMUNDA_URL=
--- a/.gitea/workflows/ci.yaml
+++ b/.gitea/workflows/ci.yaml
@@ -0,0 +1,192 @@
+# Gitea Actions CI Pipeline
+# BreakPilot Lehrer
+#
+# Services:
+#   Go: school-service, edu-search-service
+#   Python: klausur-service, backend-lehrer, agent-core
+#   Node.js: website, admin-lehrer, studio-v2
+
+name: CI
+
+on:
+  push:
+    branches: [main, develop]
+  pull_request:
+    branches: [main, develop]
+
+jobs:
+  # ========================================
+  # Lint (nur bei PRs)
+  # ========================================
+
+  go-lint:
+    runs-on: docker
+    if: github.event_name == 'pull_request'
+    container: golangci/golangci-lint:v1.55-alpine
+    steps:
+      - name: Checkout
+        run: |
+          apk add --no-cache git
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Lint Go services
+        run: |
+          for svc in school-service edu-search-service; do
+            if [ -d "$svc" ]; then
+              echo "=== Linting $svc ==="
+              cd "$svc" && golangci-lint run --timeout 5m ./... || true
+              cd ..
+            fi
+          done
+
+  python-lint:
+    runs-on: docker
+    if: github.event_name == 'pull_request'
+    container: python:3.12-slim
+    steps:
+      - name: Checkout
+        run: |
+          apt-get update -qq && apt-get install -y -qq git > /dev/null 2>&1
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Lint Python services
+        run: |
+          pip install --quiet ruff
+          for svc in backend-lehrer agent-core; do
+            if [ -d "$svc" ]; then
+              echo "=== Linting $svc ==="
+              ruff check "$svc/" --output-format=github || true
+            fi
+          done
+          if [ -d "klausur-service/backend" ]; then
+            echo "=== Linting klausur-service ==="
+            ruff check klausur-service/backend/ --output-format=github || true
+          fi
+
+  nodejs-lint:
+    runs-on: docker
+    if: github.event_name == 'pull_request'
+    container: node:20-alpine
+    steps:
+      - name: Checkout
+        run: |
+          apk add --no-cache git
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Lint Node.js services
+        run: |
+          for svc in website admin-lehrer studio-v2; do
+            if [ -d "$svc" ]; then
+              echo "=== Linting $svc ==="
+              cd "$svc"
+              npm ci --silent 2>/dev/null || npm install --silent
+              npx next lint || true
+              cd ..
+            fi
+          done
+
+  # ========================================
+  # Unit Tests
+  # ========================================
+
+  test-go-school:
+    runs-on: docker
+    container: golang:1.23-alpine
+    env:
+      CGO_ENABLED: "0"
+    steps:
+      - name: Checkout
+        run: |
+          apk add --no-cache git
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Test school-service
+        run: |
+          if [ ! -d "school-service" ]; then
+            echo "WARNUNG: school-service nicht gefunden"
+            exit 0
+          fi
+          cd school-service
+          go test -v -coverprofile=coverage.out ./... 2>&1
+          COVERAGE=$(go tool cover -func=coverage.out 2>/dev/null | tail -1 | awk '{print $3}' || echo "0%")
+          echo "Coverage: $COVERAGE"
+
+  test-go-edu-search:
+    runs-on: docker
+    container: golang:1.23-alpine
+    env:
+      CGO_ENABLED: "0"
+    steps:
+      - name: Checkout
+        run: |
+          apk add --no-cache git
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Test edu-search-service
+        run: |
+          if [ ! -d "edu-search-service" ]; then
+            echo "WARNUNG: edu-search-service nicht gefunden"
+            exit 0
+          fi
+          cd edu-search-service
+          go mod download
+          go test -v ./... 2>&1 || true
+
+  test-python-klausur:
+    runs-on: docker
+    container: python:3.12-slim
+    env:
+      CI: "true"
+    steps:
+      - name: Checkout
+        run: |
+          apt-get update -qq && apt-get install -y -qq git > /dev/null 2>&1
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Test klausur-service
+        run: |
+          if [ ! -d "klausur-service/backend" ]; then
+            echo "WARNUNG: klausur-service/backend nicht gefunden"
+            exit 0
+          fi
+          cd klausur-service/backend
+          export PYTHONPATH="$(pwd):${PYTHONPATH:-}"
+          pip install --quiet --no-cache-dir -r requirements.txt 2>/dev/null || true
+          pip install --quiet --no-cache-dir fastapi uvicorn pytest pytest-asyncio
+          python -m pytest tests/ -v --tb=short
+
+  test-python-agent-core:
+    runs-on: docker
+    container: python:3.12-slim
+    env:
+      CI: "true"
+    steps:
+      - name: Checkout
+        run: |
+          apt-get update -qq && apt-get install -y -qq git > /dev/null 2>&1
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Test agent-core
+        run: |
+          if [ ! -d "agent-core" ]; then
+            echo "WARNUNG: agent-core nicht gefunden"
+            exit 0
+          fi
+          # Symlink fuer Python import (Verzeichnis agent-core -> agent_core)
+          ln -sf agent-core agent_core
+          export PYTHONPATH="$(pwd):${PYTHONPATH:-}"
+          cd agent-core
+          pip install --quiet --no-cache-dir -r requirements.txt 2>/dev/null || true
+          pip install --quiet --no-cache-dir pytest pytest-asyncio
+          python -m pytest tests/ -v --tb=short
+
+  test-nodejs-website:
+    runs-on: docker
+    container: node:20-alpine
+    steps:
+      - name: Checkout
+        run: |
+          apk add --no-cache git
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git .
+      - name: Test website
+        run: |
+          if [ ! -d "website" ]; then
+            echo "WARNUNG: website nicht gefunden"
+            exit 0
+          fi
+          cd website
+          npm ci --silent 2>/dev/null || npm install --silent
+          npx jest --passWithNoTests || true
--- a/.gitea/workflows/deploy-coolify.yml
+++ b/.gitea/workflows/deploy-coolify.yml
@@ -0,0 +1,32 @@
+name: Deploy to Coolify
+
+on:
+  push:
+    branches:
+      - coolify
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Wait for Core deployment
+        run: |
+          echo "Waiting 30s for Core services to stabilize..."
+          sleep 30
+
+      - name: Deploy via Coolify API
+        run: |
+          echo "Deploying breakpilot-lehrer to Coolify..."
+          HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" \
+            -X POST \
+            -H "Authorization: Bearer ${{ secrets.COOLIFY_API_TOKEN }}" \
+            -H "Content-Type: application/json" \
+            -d '{"uuid": "${{ secrets.COOLIFY_RESOURCE_UUID }}", "force_rebuild": true}' \
+            "${{ secrets.COOLIFY_BASE_URL }}/api/v1/deploy")
+
+          echo "HTTP Status: $HTTP_STATUS"
+          if [ "$HTTP_STATUS" -ne 200 ] && [ "$HTTP_STATUS" -ne 201 ]; then
+            echo "Deployment failed with status $HTTP_STATUS"
+            exit 1
+          fi
+          echo "Deployment triggered successfully!"
--- a/.gitignore
+++ b/.gitignore
@@ -120,3 +120,4 @@ coverage/
 *.dll
 *.so
 *.dylib
+edu-search-service/vendor/
--- a/.woodpecker/main.yml
+++ b/.woodpecker/main.yml
@@ -4,8 +4,8 @@
 # Plattform: ARM64 (Apple Silicon Mac Mini)
 #
 # Services:
-#   Go: school-service
-#   Python: klausur-service, backend-lehrer, geo-service, agent-core
+#   Go: school-service, edu-search-service
+#   Python: klausur-service, backend-lehrer, agent-core
 #   Node.js: website, admin-lehrer, studio-v2
 #
 # Strategie:
@@ -42,9 +42,13 @@ steps:
    image: golangci/golangci-lint:v1.55-alpine
    commands:
      - |
-        if [ -d "school-service" ]; then
-          cd school-service && golangci-lint run --timeout 5m ./...
-        fi
+        for svc in school-service edu-search-service; do
+          if [ -d "$svc" ]; then
+            echo "=== Linting $svc ==="
+            cd "$svc" && golangci-lint run --timeout 5m ./... || true
+            cd ..
+          fi
+        done
    when:
      event: pull_request

@@ -53,7 +57,7 @@ steps:
    commands:
      - pip install --quiet ruff
      - |
-        for svc in backend-lehrer geo-service agent-core; do
+        for svc in backend-lehrer agent-core; do
          if [ -d "$svc" ]; then
            echo "=== Linting $svc ==="
            ruff check "$svc/" --output-format=github || true
@@ -130,6 +134,47 @@ steps:
          echo "WARNUNG: $FAILED Tests fehlgeschlagen - werden ins Backlog geschrieben"
        fi

+  test-go-edu-search:
+    image: *golang_image
+    environment:
+      CGO_ENABLED: "0"
+    commands:
+      - |
+        set -euo pipefail
+        apk add --no-cache jq bash
+        mkdir -p .ci-results
+
+        if [ ! -d "edu-search-service" ]; then
+          echo '{"service":"edu-search-service","framework":"go","total":0,"passed":0,"failed":0,"skipped":0,"coverage":0}' > .ci-results/results-edu-search.json
+          echo "WARNUNG: edu-search-service Verzeichnis nicht gefunden"
+          exit 0
+        fi
+
+        cd edu-search-service
+        go mod download
+        set +e
+        go test -v -json ./... 2>&1 | tee ../.ci-results/test-edu-search.json
+        TEST_EXIT=$?
+        set -e
+
+        JSON_FILE="../.ci-results/test-edu-search.json"
+        if grep -q '^{' "$JSON_FILE" 2>/dev/null; then
+          TOTAL=$(grep '^{' "$JSON_FILE" | jq -s '[.[] | select(.Action=="run" and .Test != null)] | length')
+          PASSED=$(grep '^{' "$JSON_FILE" | jq -s '[.[] | select(.Action=="pass" and .Test != null)] | length')
+          FAILED=$(grep '^{' "$JSON_FILE" | jq -s '[.[] | select(.Action=="fail" and .Test != null)] | length')
+          SKIPPED=$(grep '^{' "$JSON_FILE" | jq -s '[.[] | select(.Action=="skip" and .Test != null)] | length')
+        else
+          echo "WARNUNG: Keine JSON-Zeilen in $JSON_FILE gefunden (Build-Fehler?)"
+          TOTAL=0; PASSED=0; FAILED=0; SKIPPED=0
+        fi
+
+        echo "{\"service\":\"edu-search-service\",\"framework\":\"go\",\"total\":$TOTAL,\"passed\":$PASSED,\"failed\":$FAILED,\"skipped\":$SKIPPED,\"coverage\":0}" > ../.ci-results/results-edu-search.json
+        cat ../.ci-results/results-edu-search.json
+
+        if [ "$FAILED" -gt "0" ]; then
+          echo "WARNUNG: $FAILED Tests fehlgeschlagen"
+        fi
+
  test-python-klausur:
    image: *python_image
    environment:
@@ -169,52 +214,6 @@ steps:
        cat ../../.ci-results/results-klausur.json

        if [ "$TEST_EXIT" -ne "0" ]; then exit 1; fi
-
-  test-python-geo:
-    image: *python_image
-    environment:
-      CI: "true"
-    commands:
-      - |
-        set -uo pipefail
-        mkdir -p .ci-results
-
-        if [ ! -d "geo-service" ]; then
-          echo '{"service":"geo-service","framework":"pytest","total":0,"passed":0,"failed":0,"skipped":0,"coverage":0}' > .ci-results/results-geo.json
-          echo "WARNUNG: geo-service Verzeichnis nicht gefunden"
-          exit 0
-        fi
-
-        cd geo-service
-        export PYTHONPATH="$(pwd):${PYTHONPATH:-}"
-
-        # Pakete einzeln installieren - rasterio braucht GDAL und wird uebersprungen
-        while IFS= read -r line; do
-          pkg=$(echo "$line" | sed 's/#.*//' | xargs)
-          [ -z "$pkg" ] && continue
-          pip install --quiet --no-cache-dir "$pkg" 2>/dev/null || echo "WARNUNG: $pkg uebersprungen"
-        done < requirements.txt
-        pip install --quiet --no-cache-dir pytest pytest-json-report
-
-        set +e
-        python -m pytest tests/ -v --tb=short --json-report --json-report-file=../.ci-results/test-geo.json
-        TEST_EXIT=$?
-        set -e
-
-        if [ -f ../.ci-results/test-geo.json ]; then
-          TOTAL=$(python3 -c "import json; d=json.load(open('../.ci-results/test-geo.json')); print(d.get('summary',{}).get('total',0))" 2>/dev/null || echo "0")
-          PASSED=$(python3 -c "import json; d=json.load(open('../.ci-results/test-geo.json')); print(d.get('summary',{}).get('passed',0))" 2>/dev/null || echo "0")
-          FAILED=$(python3 -c "import json; d=json.load(open('../.ci-results/test-geo.json')); print(d.get('summary',{}).get('failed',0))" 2>/dev/null || echo "0")
-          SKIPPED=$(python3 -c "import json; d=json.load(open('../.ci-results/test-geo.json')); print(d.get('summary',{}).get('skipped',0))" 2>/dev/null || echo "0")
-        else
-          TOTAL=0; PASSED=0; FAILED=0; SKIPPED=0
-        fi
-
-        echo "{\"service\":\"geo-service\",\"framework\":\"pytest\",\"total\":$TOTAL,\"passed\":$PASSED,\"failed\":$FAILED,\"skipped\":$SKIPPED,\"coverage\":0}" > ../.ci-results/results-geo.json
-        cat ../.ci-results/results-geo.json
-
-        if [ "$TEST_EXIT" -ne "0" ]; then exit 1; fi
-
  test-python-agent-core:
    image: *python_image
    environment:
@@ -333,8 +332,8 @@ steps:
      status: [success, failure]
    depends_on:
      - test-go-school
+      - test-go-edu-search
      - test-python-klausur
-      - test-python-geo
      - test-python-agent-core
      - test-nodejs-website

@@ -432,16 +431,16 @@ steps:
      - event: tag
      - event: manual

-  build-geo-service:
+  build-edu-search-service:
    image: *docker_image
    commands:
      - |
-        if [ -d ./geo-service ]; then
-          docker build -t breakpilot/geo-service:${CI_COMMIT_SHA:0:8} ./geo-service
-          docker tag breakpilot/geo-service:${CI_COMMIT_SHA:0:8} breakpilot/geo-service:latest
-          echo "Built breakpilot/geo-service:${CI_COMMIT_SHA:0:8}"
+        if [ -d ./edu-search-service ]; then
+          docker build -t breakpilot/edu-search-service:${CI_COMMIT_SHA:0:8} ./edu-search-service
+          docker tag breakpilot/edu-search-service:${CI_COMMIT_SHA:0:8} breakpilot/edu-search-service:latest
+          echo "Built breakpilot/edu-search-service:${CI_COMMIT_SHA:0:8}"
        else
-          echo "geo-service Verzeichnis nicht gefunden - ueberspringe"
+          echo "edu-search-service Verzeichnis nicht gefunden - ueberspringe"
        fi
    when:
      - event: tag
@@ -454,7 +453,7 @@ steps:
        echo "Installing syft for ARM64..."
        apt-get update -qq && apt-get install -y -qq wget > /dev/null
        wget -qO- https://raw.githubusercontent.com/anchore/syft/main/install.sh | sh -s -- -b /usr/local/bin
-        for svc in klausur-service backend-lehrer website school-service geo-service agent-core; do
+        for svc in klausur-service backend-lehrer website school-service edu-search-service agent-core; do
          if [ -d "./$svc" ]; then
            syft dir:./$svc -o cyclonedx-json > sbom-$svc.json
            echo "SBOM generated for $svc"
@@ -501,4 +500,4 @@ steps:
      - build-backend-lehrer
      - build-klausur-service
      - build-school-service
-      - build-geo-service
+      - build-edu-search-service
--- a/admin-lehrer/Dockerfile
+++ b/admin-lehrer/Dockerfile
@@ -34,8 +34,8 @@ WORKDIR /app
 ENV NODE_ENV=production

 # Create non-root user
-RUN addgroup --system --gid 1001 nodejs
-RUN adduser --system --uid 1001 nextjs
+RUN addgroup -S -g 1001 nodejs
+RUN adduser -S -u 1001 -G nodejs nextjs

 # Copy built assets
 COPY --from=builder /app/public ./public
--- a/docker-compose.coolify.yml
+++ b/docker-compose.coolify.yml
@@ -0,0 +1,323 @@
+# =========================================================
+# BreakPilot Lehrer — KI-Lehrerplattform (Coolify)
+# =========================================================
+# Requires: breakpilot-core must be running
+# Deployed via Coolify. SSL termination handled by Traefik.
+# External services (managed separately in Coolify):
+#   - PostgreSQL, Qdrant, S3-compatible storage
+# =========================================================
+
+networks:
+  breakpilot-network:
+    external: true
+    name: breakpilot-network
+
+volumes:
+  klausur_uploads:
+  eh_uploads:
+  ocr_labeling:
+  paddle_models:
+  lehrer_backend_data:
+  opensearch_data:
+
+services:
+
+  # =========================================================
+  # FRONTEND
+  # =========================================================
+  admin-lehrer:
+    build:
+      context: ./admin-lehrer
+      dockerfile: Dockerfile
+      args:
+        NEXT_PUBLIC_API_URL: ${NEXT_PUBLIC_API_URL:-https://api-lehrer.breakpilot.ai}
+        NEXT_PUBLIC_OLD_ADMIN_URL: ${NEXT_PUBLIC_OLD_ADMIN_URL:-}
+        NEXT_PUBLIC_KLAUSUR_SERVICE_URL: ${NEXT_PUBLIC_KLAUSUR_SERVICE_URL:-https://klausur.breakpilot.ai}
+        NEXT_PUBLIC_VOICE_SERVICE_URL: ${NEXT_PUBLIC_VOICE_SERVICE_URL:-wss://voice.breakpilot.ai}
+    container_name: bp-lehrer-admin
+    expose:
+      - "3000"
+    volumes:
+      - lehrer_backend_data:/app/data
+    environment:
+      NODE_ENV: production
+      BACKEND_URL: http://backend-lehrer:8001
+      CONSENT_SERVICE_URL: http://bp-core-consent-service:8081
+      KLAUSUR_SERVICE_URL: http://klausur-service:8086
+      OLLAMA_URL: ${OLLAMA_URL:-}
+    depends_on:
+      backend-lehrer:
+        condition: service_started
+    labels:
+      - "traefik.enable=true"
+      - "traefik.http.routers.admin-lehrer.rule=Host(`admin-lehrer.breakpilot.ai`)"
+      - "traefik.http.routers.admin-lehrer.entrypoints=https"
+      - "traefik.http.routers.admin-lehrer.tls=true"
+      - "traefik.http.routers.admin-lehrer.tls.certresolver=letsencrypt"
+      - "traefik.http.services.admin-lehrer.loadbalancer.server.port=3000"
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  studio-v2:
+    build:
+      context: ./studio-v2
+      dockerfile: Dockerfile
+      args:
+        NEXT_PUBLIC_VOICE_SERVICE_URL: ${NEXT_PUBLIC_VOICE_SERVICE_URL:-wss://voice.breakpilot.ai}
+        NEXT_PUBLIC_KLAUSUR_SERVICE_URL: ${NEXT_PUBLIC_KLAUSUR_SERVICE_URL:-https://klausur.breakpilot.ai}
+    container_name: bp-lehrer-studio-v2
+    expose:
+      - "3001"
+    environment:
+      NODE_ENV: production
+      BACKEND_URL: http://backend-lehrer:8001
+    depends_on:
+      - backend-lehrer
+    labels:
+      - "traefik.enable=true"
+      - "traefik.http.routers.studio.rule=Host(`app.breakpilot.ai`)"
+      - "traefik.http.routers.studio.entrypoints=https"
+      - "traefik.http.routers.studio.tls=true"
+      - "traefik.http.routers.studio.tls.certresolver=letsencrypt"
+      - "traefik.http.services.studio.loadbalancer.server.port=3001"
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  website:
+    build:
+      context: ./website
+      dockerfile: Dockerfile
+      args:
+        NEXT_PUBLIC_BILLING_API_URL: ${NEXT_PUBLIC_BILLING_API_URL:-https://api-core.breakpilot.ai}
+        NEXT_PUBLIC_APP_URL: ${NEXT_PUBLIC_APP_URL:-https://app.breakpilot.ai}
+        NEXT_PUBLIC_KLAUSUR_SERVICE_URL: ${NEXT_PUBLIC_KLAUSUR_SERVICE_URL:-https://klausur.breakpilot.ai}
+        NEXT_PUBLIC_VOICE_SERVICE_URL: ${NEXT_PUBLIC_VOICE_SERVICE_URL:-wss://voice.breakpilot.ai}
+        NEXT_PUBLIC_STRIPE_PUBLISHABLE_KEY: ${NEXT_PUBLIC_STRIPE_PUBLISHABLE_KEY:-}
+    container_name: bp-lehrer-website
+    expose:
+      - "3000"
+    environment:
+      NODE_ENV: production
+      VAST_API_KEY: ${VAST_API_KEY:-}
+      CONTROL_API_KEY: ${CONTROL_API_KEY:-}
+      BACKEND_URL: http://backend-lehrer:8001
+      CONSENT_SERVICE_URL: http://bp-core-consent-service:8081
+      EDU_SEARCH_URL: ${EDU_SEARCH_URL:-}
+      EDU_SEARCH_API_KEY: ${EDU_SEARCH_API_KEY:-}
+    depends_on:
+      - backend-lehrer
+    labels:
+      - "traefik.enable=true"
+      - "traefik.http.routers.website.rule=Host(`www.breakpilot.ai`)"
+      - "traefik.http.routers.website.entrypoints=https"
+      - "traefik.http.routers.website.tls=true"
+      - "traefik.http.routers.website.tls.certresolver=letsencrypt"
+      - "traefik.http.services.website.loadbalancer.server.port=3000"
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  # =========================================================
+  # BACKEND
+  # =========================================================
+  backend-lehrer:
+    build:
+      context: ./backend-lehrer
+      dockerfile: Dockerfile
+    container_name: bp-lehrer-backend
+    user: "0:0"
+    expose:
+      - "8001"
+    volumes:
+      - lehrer_backend_data:/app/data
+    environment:
+      PORT: 8001
+      DATABASE_URL: postgresql+asyncpg://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT:-5432}/${POSTGRES_DB}?options=-csearch_path%3Dlehrer,core,public
+      JWT_SECRET: ${JWT_SECRET}
+      ENVIRONMENT: production
+      CONSENT_SERVICE_URL: http://bp-core-consent-service:8081
+      KLAUSUR_SERVICE_URL: http://klausur-service:8086
+      TROCR_SERVICE_URL: ${TROCR_SERVICE_URL:-}
+      CAMUNDA_URL: ${CAMUNDA_URL:-}
+      VALKEY_URL: redis://bp-core-valkey:6379/0
+      SESSION_TTL_HOURS: ${SESSION_TTL_HOURS:-24}
+      ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY:-}
+      DEBUG: "false"
+      ALERTS_AGENT_ENABLED: ${ALERTS_AGENT_ENABLED:-false}
+      VAST_API_KEY: ${VAST_API_KEY:-}
+      VAST_INSTANCE_ID: ${VAST_INSTANCE_ID:-}
+      CONTROL_API_KEY: ${CONTROL_API_KEY:-}
+      OLLAMA_BASE_URL: ${OLLAMA_BASE_URL:-}
+      OLLAMA_ENABLED: ${OLLAMA_ENABLED:-false}
+      OLLAMA_DEFAULT_MODEL: ${OLLAMA_DEFAULT_MODEL:-}
+      OLLAMA_VISION_MODEL: ${OLLAMA_VISION_MODEL:-}
+      OLLAMA_CORRECTION_MODEL: ${OLLAMA_CORRECTION_MODEL:-}
+      OLLAMA_TIMEOUT: ${OLLAMA_TIMEOUT:-120}
+      GAME_USE_DATABASE: ${GAME_USE_DATABASE:-true}
+      GAME_REQUIRE_AUTH: ${GAME_REQUIRE_AUTH:-true}
+      GAME_REQUIRE_BILLING: ${GAME_REQUIRE_BILLING:-true}
+      GAME_LLM_MODEL: ${GAME_LLM_MODEL:-}
+      SMTP_HOST: ${SMTP_HOST}
+      SMTP_PORT: ${SMTP_PORT:-587}
+      SMTP_USERNAME: ${SMTP_USERNAME}
+      SMTP_PASSWORD: ${SMTP_PASSWORD}
+      SMTP_FROM_NAME: ${SMTP_FROM_NAME:-BreakPilot}
+      SMTP_FROM_ADDR: ${SMTP_FROM_ADDR:-noreply@breakpilot.ai}
+      RAG_SERVICE_URL: http://bp-core-rag-service:8097
+    labels:
+      - "traefik.enable=true"
+      - "traefik.http.routers.backend-lehrer.rule=Host(`api-lehrer.breakpilot.ai`)"
+      - "traefik.http.routers.backend-lehrer.entrypoints=https"
+      - "traefik.http.routers.backend-lehrer.tls=true"
+      - "traefik.http.routers.backend-lehrer.tls.certresolver=letsencrypt"
+      - "traefik.http.services.backend-lehrer.loadbalancer.server.port=8001"
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  # =========================================================
+  # MICROSERVICES
+  # =========================================================
+  klausur-service:
+    build:
+      context: ./klausur-service
+      dockerfile: Dockerfile
+    container_name: bp-lehrer-klausur-service
+    expose:
+      - "8086"
+    volumes:
+      - klausur_uploads:/app/uploads
+      - eh_uploads:/app/eh-uploads
+      - ocr_labeling:/app/ocr-labeling
+      - paddle_models:/root/.paddlex
+    environment:
+      JWT_SECRET: ${JWT_SECRET}
+      BACKEND_URL: http://backend-lehrer:8001
+      SCHOOL_SERVICE_URL: http://school-service:8084
+      ENVIRONMENT: production
+      DATABASE_URL: postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT:-5432}/${POSTGRES_DB}
+      EMBEDDING_SERVICE_URL: http://bp-core-embedding-service:8087
+      QDRANT_URL: ${QDRANT_URL}
+      MINIO_ENDPOINT: ${S3_ENDPOINT}
+      MINIO_ACCESS_KEY: ${S3_ACCESS_KEY}
+      MINIO_SECRET_KEY: ${S3_SECRET_KEY}
+      MINIO_BUCKET: ${S3_BUCKET:-breakpilot-rag}
+      MINIO_SECURE: ${S3_SECURE:-true}
+      PADDLEOCR_SERVICE_URL: ${PADDLEOCR_SERVICE_URL:-}
+      ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY:-}
+      OLLAMA_BASE_URL: ${OLLAMA_BASE_URL:-}
+      OLLAMA_ENABLED: ${OLLAMA_ENABLED:-false}
+      OLLAMA_DEFAULT_MODEL: ${OLLAMA_DEFAULT_MODEL:-}
+      OLLAMA_VISION_MODEL: ${OLLAMA_VISION_MODEL:-}
+      OLLAMA_CORRECTION_MODEL: ${OLLAMA_CORRECTION_MODEL:-}
+      RAG_SERVICE_URL: http://bp-core-rag-service:8097
+    depends_on:
+      school-service:
+        condition: service_started
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://127.0.0.1:8086/health"]
+      interval: 30s
+      timeout: 30s
+      retries: 3
+      start_period: 10s
+    labels:
+      - "traefik.enable=true"
+      - "traefik.http.routers.klausur.rule=Host(`klausur.breakpilot.ai`)"
+      - "traefik.http.routers.klausur.entrypoints=https"
+      - "traefik.http.routers.klausur.tls=true"
+      - "traefik.http.routers.klausur.tls.certresolver=letsencrypt"
+      - "traefik.http.services.klausur.loadbalancer.server.port=8086"
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  school-service:
+    build:
+      context: ./school-service
+      dockerfile: Dockerfile
+    container_name: bp-lehrer-school-service
+    expose:
+      - "8084"
+    environment:
+      DATABASE_URL: postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT:-5432}/${POSTGRES_DB}
+      JWT_SECRET: ${JWT_SECRET}
+      PORT: 8084
+      ENVIRONMENT: production
+      ALLOWED_ORIGINS: "*"
+      LLM_GATEWAY_URL: http://backend-lehrer:8001/llm
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  # =========================================================
+  # EDU SEARCH
+  # =========================================================
+  opensearch:
+    image: opensearchproject/opensearch:2.11.1
+    container_name: bp-lehrer-opensearch
+    environment:
+      - cluster.name=edu-search-cluster
+      - node.name=opensearch-node1
+      - discovery.type=single-node
+      - bootstrap.memory_lock=true
+      - "OPENSEARCH_JAVA_OPTS=-Xms512m -Xmx512m"
+      - OPENSEARCH_INITIAL_ADMIN_PASSWORD=${OPENSEARCH_PASSWORD:-Admin123!}
+      - plugins.security.disabled=true
+    ulimits:
+      memlock:
+        soft: -1
+        hard: -1
+      nofile:
+        soft: 65536
+        hard: 65536
+    volumes:
+      - opensearch_data:/usr/share/opensearch/data
+    healthcheck:
+      test: ["CMD-SHELL", "curl -s http://localhost:9200 >/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  edu-search-service:
+    build:
+      context: ./edu-search-service
+      dockerfile: Dockerfile
+    container_name: bp-lehrer-edu-search
+    expose:
+      - "8088"
+    environment:
+      PORT: 8088
+      OPENSEARCH_URL: http://opensearch:9200
+      OPENSEARCH_USERNAME: admin
+      OPENSEARCH_PASSWORD: ${OPENSEARCH_PASSWORD:-Admin123!}
+      INDEX_NAME: bp_documents_v1
+      EDU_SEARCH_API_KEY: ${EDU_SEARCH_API_KEY:-}
+      USER_AGENT: "BreakpilotEduCrawler/1.0 (+contact: security@breakpilot.com)"
+      RATE_LIMIT_PER_SEC: "0.2"
+      MAX_DEPTH: "4"
+      MAX_PAGES_PER_RUN: "500"
+      DB_HOST: ${POSTGRES_HOST}
+      DB_PORT: ${POSTGRES_PORT:-5432}
+      DB_USER: ${POSTGRES_USER}
+      DB_PASSWORD: ${POSTGRES_PASSWORD}
+      DB_NAME: ${POSTGRES_DB}
+      DB_SSLMODE: disable
+      STAFF_CRAWLER_EMAIL: crawler@breakpilot.de
+    depends_on:
+      opensearch:
+        condition: service_healthy
+    healthcheck:
+      test: ["CMD", "wget", "--no-verbose", "--tries=1", "--spider", "http://localhost:8088/v1/health"]
+      interval: 30s
+      timeout: 3s
+      start_period: 10s
+      retries: 3
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -16,14 +16,10 @@ volumes:
  ocr_labeling:
  paddle_models:
  paddleocr_models:
-  voice_session_data:
-  geo_osm_data:
-  geo_dem_data:
-  geo_tile_cache:
-  geo_aoi_bundles:
  transcription_models:
  transcription_temp:
  lehrer_backend_data:
+  opensearch_data:

 services:

@@ -275,83 +271,6 @@ services:
    networks:
      - breakpilot-network

-  geo-service:
-    build:
-      context: ./geo-service
-      dockerfile: Dockerfile
-    container_name: bp-lehrer-geo-service
-    platform: linux/arm64
-    ports:
-      - "8088:8088"
-    volumes:
-      - geo_osm_data:/app/data/osm
-      - geo_dem_data:/app/data/dem
-      - geo_tile_cache:/app/cache/tiles
-      - geo_aoi_bundles:/app/bundles
-    environment:
-      PORT: 8088
-      ENVIRONMENT: ${ENVIRONMENT:-development}
-      JWT_SECRET: ${JWT_SECRET:-your-super-secret-jwt-key-change-in-production}
-      DATABASE_URL: postgresql://${POSTGRES_USER:-breakpilot}:${POSTGRES_PASSWORD:-breakpilot123}@bp-core-postgres:5432/${POSTGRES_DB:-breakpilot_db}
-      MINIO_ENDPOINT: bp-core-minio:9000
-      MINIO_ACCESS_KEY: ${MINIO_ROOT_USER:-breakpilot}
-      MINIO_SECRET_KEY: ${MINIO_ROOT_PASSWORD:-breakpilot123}
-      MINIO_BUCKET: ${MINIO_BUCKET:-breakpilot-geo}
-      MINIO_SECURE: "false"
-      OLLAMA_BASE_URL: ${OLLAMA_BASE_URL:-http://host.docker.internal:11434}
-      OLLAMA_MODEL: ${OLLAMA_DEFAULT_MODEL:-llama3.2}
-      TILE_CACHE_DIR: /app/cache/tiles
-      DEM_CACHE_DIR: /app/data/dem
-      MAX_AOI_SIZE_KM2: ${MAX_AOI_SIZE_KM2:-100}
-    extra_hosts:
-      - "host.docker.internal:host-gateway"
-    depends_on:
-      core-health-check:
-        condition: service_completed_successfully
-    healthcheck:
-      test: ["CMD", "curl", "-f", "http://127.0.0.1:8088/health"]
-      interval: 30s
-      timeout: 10s
-      start_period: 60s
-      retries: 3
-    restart: unless-stopped
-    networks:
-      - breakpilot-network
-
-  voice-service:
-    build:
-      context: ./voice-service
-      dockerfile: Dockerfile
-    container_name: bp-lehrer-voice-service
-    platform: linux/arm64
-    expose:
-      - "8091"
-    volumes:
-      - voice_session_data:/app/data/sessions
-    environment:
-      PORT: 8091
-      DATABASE_URL: postgresql://${POSTGRES_USER:-breakpilot}:${POSTGRES_PASSWORD:-breakpilot123}@bp-core-postgres:5432/${POSTGRES_DB:-breakpilot_db}
-      VALKEY_URL: redis://bp-core-valkey:6379/0
-      KLAUSUR_SERVICE_URL: http://klausur-service:8086
-      OLLAMA_BASE_URL: ${OLLAMA_BASE_URL:-http://host.docker.internal:11434}
-      OLLAMA_VOICE_MODEL: ${OLLAMA_VOICE_MODEL:-llama3.2}
-      ENVIRONMENT: ${ENVIRONMENT:-development}
-      JWT_SECRET: ${JWT_SECRET:-your-super-secret-jwt-key-change-in-production}
-    extra_hosts:
-      - "host.docker.internal:host-gateway"
-    depends_on:
-      core-health-check:
-        condition: service_completed_successfully
-    healthcheck:
-      test: ["CMD", "curl", "-f", "http://127.0.0.1:8091/health"]
-      interval: 30s
-      timeout: 10s
-      start_period: 60s
-      retries: 3
-    restart: unless-stopped
-    networks:
-      - breakpilot-network
-
  paddleocr-service:
    build:
      context: ./paddleocr-service
@@ -454,6 +373,80 @@ services:
    networks:
      - breakpilot-network

+  # =========================================================
+  # EDU SEARCH
+  # =========================================================
+  opensearch:
+    image: opensearchproject/opensearch:2.11.1
+    container_name: bp-lehrer-opensearch
+    environment:
+      - cluster.name=edu-search-cluster
+      - node.name=opensearch-node1
+      - discovery.type=single-node
+      - bootstrap.memory_lock=true
+      - "OPENSEARCH_JAVA_OPTS=-Xms512m -Xmx512m"
+      - OPENSEARCH_INITIAL_ADMIN_PASSWORD=${OPENSEARCH_PASSWORD:-Admin123!}
+      - plugins.security.disabled=true
+    ulimits:
+      memlock:
+        soft: -1
+        hard: -1
+      nofile:
+        soft: 65536
+        hard: 65536
+    volumes:
+      - opensearch_data:/usr/share/opensearch/data
+    healthcheck:
+      test: ["CMD-SHELL", "curl -s http://localhost:9200 >/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
+  edu-search-service:
+    build:
+      context: ./edu-search-service
+      dockerfile: Dockerfile
+    container_name: bp-lehrer-edu-search
+    platform: linux/arm64
+    expose:
+      - "8088"
+    environment:
+      PORT: 8088
+      OPENSEARCH_URL: http://opensearch:9200
+      OPENSEARCH_USERNAME: admin
+      OPENSEARCH_PASSWORD: ${OPENSEARCH_PASSWORD:-Admin123!}
+      INDEX_NAME: bp_documents_v1
+      EDU_SEARCH_API_KEY: ${EDU_SEARCH_API_KEY:-}
+      USER_AGENT: "BreakpilotEduCrawler/1.0 (+contact: security@breakpilot.com)"
+      RATE_LIMIT_PER_SEC: "0.2"
+      MAX_DEPTH: "4"
+      MAX_PAGES_PER_RUN: "500"
+      DB_HOST: bp-core-postgres
+      DB_PORT: "5432"
+      DB_USER: ${POSTGRES_USER:-breakpilot}
+      DB_PASSWORD: ${POSTGRES_PASSWORD:-breakpilot123}
+      DB_NAME: ${POSTGRES_DB:-breakpilot_db}
+      DB_SSLMODE: disable
+      STAFF_CRAWLER_EMAIL: crawler@breakpilot.de
+    depends_on:
+      core-health-check:
+        condition: service_completed_successfully
+      opensearch:
+        condition: service_healthy
+    healthcheck:
+      test: ["CMD", "wget", "--no-verbose", "--tries=1", "--spider", "http://localhost:8088/v1/health"]
+      interval: 30s
+      timeout: 3s
+      start_period: 10s
+      retries: 3
+    restart: unless-stopped
+    networks:
+      - breakpilot-network
+
  # =========================================================
  # DOCUMENTATION
  # =========================================================
--- a/edu-search-service/Dockerfile
+++ b/edu-search-service/Dockerfile
@@ -0,0 +1,48 @@
+# Build stage
+FROM golang:1.23-alpine AS builder
+
+WORKDIR /app
+
+# Copy go mod files and vendor
+COPY go.mod go.sum ./
+COPY vendor/ vendor/
+
+# Copy source code
+COPY . .
+
+# Build binary with vendor mode
+RUN CGO_ENABLED=0 GOOS=linux go build -mod=vendor -a -installsuffix cgo -o edu-search-service ./cmd/server
+
+# Runtime stage
+FROM alpine:3.19
+
+WORKDIR /app
+
+# Install CA certificates for HTTPS
+RUN apk --no-cache add ca-certificates tzdata
+
+# Create non-root user
+RUN adduser -D -g '' appuser
+
+# Copy binary from builder
+COPY --from=builder /app/edu-search-service .
+
+# Copy seeds, rules and migrations
+COPY seeds/ ./seeds/
+COPY rules/ ./rules/
+COPY migrations/ ./migrations/
+
+# Set ownership
+RUN chown -R appuser:appuser /app
+
+USER appuser
+
+# Expose port
+EXPOSE 8086
+
+# Health check
+HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \
+    CMD wget --no-verbose --tries=1 --spider http://localhost:8086/v1/health || exit 1
+
+# Run
+CMD ["./edu-search-service"]
--- a/edu-search-service/README.md
+++ b/edu-search-service/README.md
@@ -0,0 +1,409 @@
+# edu-search-service
+
+Spezialisierter Suchdienst für deutsche Bildungsinhalte - eine Alternative zu Tavily, optimiert für den deutschen Bildungssektor.
+
+## Übersicht
+
+Der edu-search-service crawlt, extrahiert und indiziert Bildungsinhalte von deutschen Bildungsquellen (Kultusministerien, Bildungsserver, wissenschaftliche Studien, etc.) und stellt eine Such-API bereit.
+
+### Features
+
+- **BM25 Keyword-Suche** mit German Analyzer (OpenSearch)
+- **Semantic Search** mit Embeddings (OpenAI oder Ollama)
+- **Hybrid Search** kombiniert BM25 + Vektor-Ähnlichkeit
+- **Automatisches Tagging** für Dokumenttyp, Fächer, Schulstufe, Bundesland
+- **Trust-Score** basierend auf Domain-Reputation und Content-Qualität
+- **Rate-Limited Crawler** mit robots.txt Respekt
+- **Admin API** für Seed-Verwaltung und Crawl-Steuerung
+
+## Architektur
+
+```
+┌─────────────────────────────────────────────────────────────────────┐
+│                        edu-search-service                           │
+├─────────────────────────────────────────────────────────────────────┤
+│                                                                     │
+│   ┌─────────┐    ┌───────────┐    ┌────────┐    ┌─────────┐       │
+│   │ Crawler │───▶│ Extractor │───▶│ Tagger │───▶│ Indexer │       │
+│   └─────────┘    └───────────┘    └────────┘    └─────────┘       │
+│        │                                              │             │
+│        ▼                                              ▼             │
+│   ┌─────────┐                                  ┌────────────┐      │
+│   │  Seeds  │                                  │ OpenSearch │      │
+│   └─────────┘                                  └────────────┘      │
+│                                                      │             │
+│                     ┌────────────┐                   │             │
+│                     │ Search API │◀──────────────────┘             │
+│                     └────────────┘                                 │
+│                                                                     │
+└─────────────────────────────────────────────────────────────────────┘
+```
+
+## Komponenten
+
+### Crawler (`internal/crawler/`)
+- Rate-Limited HTTP Client (Standard: 0.2 req/sec pro Domain)
+- Denylist-Support für ungewünschte Domains
+- **Seeds aus Backend-API** (primär) oder lokale Seed-Files (Fallback)
+- URL-Normalisierung und Deduplication
+- Seed-Metadaten: Trust-Boost, Crawl-Tiefe, Kategorie, Bundesland
+- **Crawl-Status-Feedback** an Backend (Dokumentenzahl, Dauer, Fehler)
+
+### Robots (`internal/robots/`)
+- **robots.txt Parser** mit Caching (24h TTL)
+- Unterstützt `Disallow`, `Allow`, `Crawl-delay`
+- Wildcard-Patterns (`*`) und End-Anchors (`$`)
+- User-Agent-spezifische Regeln
+- Leniente Behandlung bei fehlenden robots.txt
+
+### Extractor (`internal/extractor/`)
+- HTML-Extraktion mit goquery
+- **PDF-Textextraktion** mit ledongthuc/pdf Bibliothek
+  - `ExtractPDF()` - Standard-Extraktion mit GetPlainText
+  - `ExtractPDFWithMetadata()` - Seiten-weise Extraktion für mehr Kontrolle
+  - Fallback-Extraktion bei beschädigten PDFs
+  - Automatische Titel-Erkennung (erste signifikante Zeile)
+  - Heading-Erkennung (All-Caps, nummerierte Zeilen)
+- Metadaten-Extraktion (og:title, description, etc.)
+- Content-Feature-Berechnung (Ad-Density, Link-Density)
+- Sprach-Erkennung (Deutsch/Englisch)
+
+### Tagger (`internal/tagger/`)
+- Regelbasiertes Tagging via YAML-Konfiguration
+- DocType-Erkennung (Lehrplan, Arbeitsblatt, Studie, etc.)
+- Fächer-Erkennung (Mathematik, Deutsch, etc.)
+- Schulstufen-Erkennung (Grundschule, Sek I/II, etc.)
+- Bundesland-Erkennung aus URL-Patterns
+- Trust-Score-Berechnung
+
+### Quality (`internal/quality/`)
+- **Multi-Faktor Quality-Score** (0-1)
+  - Content Length (20%)
+  - Heading Structure (15%)
+  - Link/Ad Quality (15%)
+  - Text-to-HTML Ratio (15%)
+  - Metadata Presence (10%)
+  - Language Clarity (10%)
+  - Content Freshness (10%)
+  - PDF-Specific Signals (5%)
+- Konfigurierbare Gewichtungen
+- Date-Indicator-Extraktion für Frische-Bewertung
+
+### Indexer (`internal/indexer/`)
+- OpenSearch 2.11 Client
+- German Analyzer für BM25
+- Bulk-Indexierung
+- Custom Mapping für Bildungsdokumente
+
+### Search (`internal/search/`)
+- Multi-Match Query mit Boosting
+- Filter für alle Taxonomie-Felder
+- Function-Score mit Trust/Quality-Boosting
+- Highlighting-Support
+- **Drei Suchmodi:**
+  - `keyword` - Klassische BM25-Suche (Default)
+  - `semantic` - Reine Vektor-Ähnlichkeitssuche (k-NN)
+  - `hybrid` - Kombination aus BM25 und Vektor-Score
+
+### Embedding (`internal/embedding/`)
+- **OpenAI Provider** - `text-embedding-3-small` (1536 Dimensionen)
+- **Ollama Provider** - Lokale Modelle (z.B. `nomic-embed-text`, 384-768 Dim.)
+- Batch-Embedding für effiziente Indexierung
+- Automatische Text-Kürzung (max. 30.000 Zeichen)
+
+### Scheduler (`internal/scheduler/`)
+- **Automatisches Crawling** in konfigurierbaren Intervallen
+- Default: täglich um 2:00 Uhr (minimale Auswirkung)
+- Manuelles Triggern via Admin-API
+- Status-Tracking (letzter Lauf, nächster Lauf, Ergebnis)
+
+## API Endpoints
+
+### Public Endpoints
+
+| Method | Endpoint | Beschreibung |
+|--------|----------|--------------|
+| GET | `/v1/health` | Health Check (kein Auth) |
+| POST | `/v1/search` | Suche ausführen |
+| GET | `/v1/document` | Einzeldokument abrufen |
+
+### Admin Endpoints (Auth erforderlich)
+
+| Method | Endpoint | Beschreibung |
+|--------|----------|--------------|
+| GET | `/v1/admin/seeds` | Alle Seeds abrufen |
+| POST | `/v1/admin/seeds` | Neuen Seed erstellen |
+| PUT | `/v1/admin/seeds/:id` | Seed aktualisieren |
+| DELETE | `/v1/admin/seeds/:id` | Seed löschen |
+| GET | `/v1/admin/stats` | Crawl-Statistiken |
+| POST | `/v1/admin/crawl/start` | Crawl starten |
+
+## API Dokumentation
+
+### POST /v1/search
+
+**Request Body:**
+```json
+{
+  "q": "Lehrplan Mathematik Gymnasium",
+  "mode": "keyword",
+  "limit": 10,
+  "offset": 0,
+  "filters": {
+    "language": ["de"],
+    "doc_type": ["Lehrplan"],
+    "school_level": ["Gymnasium"],
+    "state": ["BY", "NW"],
+    "subjects": ["Mathematik"],
+    "min_trust_score": 0.5
+  },
+  "include": {
+    "snippets": true,
+    "highlights": true
+  }
+}
+```
+
+**Such-Modi (`mode`):**
+| Mode | Beschreibung |
+|------|--------------|
+| `keyword` | BM25-Textsuche (Default) |
+| `semantic` | Vektor-Ähnlichkeitssuche via Embeddings |
+| `hybrid` | Kombination: 70% BM25 + 30% Vektor-Score |
+
+> **Hinweis:** `semantic` und `hybrid` Modi erfordern `SEMANTIC_SEARCH_ENABLED=true` und konfigurierte Embedding-Provider.
+
+**Response:**
+```json
+{
+  "query_id": "q-12345",
+  "results": [
+    {
+      "doc_id": "uuid-...",
+      "title": "Lehrplan Mathematik Gymnasium Bayern",
+      "url": "https://www.isb.bayern.de/...",
+      "domain": "isb.bayern.de",
+      "language": "de",
+      "doc_type": "Lehrplan",
+      "school_level": "Gymnasium",
+      "subjects": ["Mathematik"],
+      "scores": {
+        "bm25": 12.5,
+        "trust": 0.85,
+        "quality": 0.9,
+        "final": 10.6
+      },
+      "snippet": "Der Lehrplan für das Fach Mathematik...",
+      "highlights": ["<em>Lehrplan</em> für das Fach <em>Mathematik</em>..."]
+    }
+  ],
+  "pagination": {
+    "limit": 10,
+    "offset": 0,
+    "total_estimate": 156
+  }
+}
+```
+
+### Filter-Optionen
+
+| Filter | Werte |
+|--------|-------|
+| `language` | `de`, `en` |
+| `doc_type` | `Lehrplan`, `Arbeitsblatt`, `Unterrichtsentwurf`, `Erlass_Verordnung`, `Pruefung_Abitur`, `Studie_Bericht`, `Sonstiges` |
+| `school_level` | `Grundschule`, `Sek_I`, `Gymnasium`, `Berufsschule`, `Hochschule`, `Alle`, `NA` |
+| `state` | `BW`, `BY`, `BE`, `BB`, `HB`, `HH`, `HE`, `MV`, `NI`, `NW`, `RP`, `SL`, `SN`, `ST`, `SH`, `TH` |
+| `subjects` | `Mathematik`, `Deutsch`, `Englisch`, `Geschichte`, `Physik`, `Biologie`, `Chemie`, etc. |
+
+## Konfiguration
+
+### Umgebungsvariablen
+
+| Variable | Beschreibung | Default |
+|----------|--------------|---------|
+| `PORT` | Server Port | `8084` |
+| `OPENSEARCH_URL` | OpenSearch URL | `http://opensearch:9200` |
+| `OPENSEARCH_USERNAME` | OpenSearch User | `admin` |
+| `OPENSEARCH_PASSWORD` | OpenSearch Passwort | `admin` |
+| `INDEX_NAME` | Index Name | `bp_documents_v1` |
+| `USER_AGENT` | Crawler User Agent | `BreakpilotEduCrawler/1.0` |
+| `RATE_LIMIT_PER_SEC` | Requests pro Sekunde/Domain | `0.2` |
+| `MAX_DEPTH` | Max Crawl-Tiefe | `4` |
+| `MAX_PAGES_PER_RUN` | Max Seiten pro Crawl | `500` |
+| `SEEDS_DIR` | Seed-Dateien Verzeichnis | `./seeds` |
+| `RULES_DIR` | Tagging-Regeln Verzeichnis | `./rules` |
+| `EDU_SEARCH_API_KEY` | API Key für Auth | `` |
+| `BACKEND_URL` | URL zum Python Backend | `http://backend:8000` |
+| `SEEDS_FROM_API` | Seeds aus API laden | `true` |
+| **Semantic Search** | | |
+| `SEMANTIC_SEARCH_ENABLED` | Semantic Search aktivieren | `false` |
+| `EMBEDDING_PROVIDER` | Provider: `openai`, `ollama`, `none` | `none` |
+| `OPENAI_API_KEY` | API Key für OpenAI Embeddings | `` |
+| `EMBEDDING_MODEL` | Embedding-Modell | `text-embedding-3-small` |
+| `EMBEDDING_DIMENSION` | Vektor-Dimension | `1536` |
+| `OLLAMA_URL` | Ollama Server URL | `http://ollama:11434` |
+| **Scheduler** | | |
+| `SCHEDULER_ENABLED` | Automatisches Crawling aktivieren | `false` |
+| `SCHEDULER_INTERVAL` | Crawl-Intervall | `24h` (täglich) |
+
+## Installation & Start
+
+### Docker (empfohlen)
+
+```bash
+# Im edu-search-service Verzeichnis
+docker compose up -d
+
+# Logs anzeigen
+docker compose logs -f edu-search
+
+# Nur der Service (OpenSearch extern)
+docker build -t edu-search-service .
+docker run -p 8084:8084 \
+  -e OPENSEARCH_URL=http://host.docker.internal:9200 \
+  edu-search-service
+```
+
+### Lokal (Entwicklung)
+
+```bash
+# Dependencies installieren
+go mod download
+
+# Service starten
+go run cmd/server/main.go
+
+# Tests ausführen
+go test -v ./...
+```
+
+## Seed-Kategorien
+
+| Kategorie | Beschreibung | Beispiele |
+|-----------|--------------|-----------|
+| `federal` | Bundesweite Institutionen | KMK, BMBF, IQB |
+| `states` | Landeskultusbehörden | Kultusministerien, Landesinstitute |
+| `science` | Wissenschaftliche Studien | PISA, IGLU, TIMSS |
+| `universities` | Hochschulen | Pädagogische Hochschulen |
+| `schools` | Schulen direkt | Schulhomepages |
+| `portals` | Bildungsportale | Lehrer-Online, 4teachers |
+| `eu` | EU-Bildungsprogramme | Erasmus+, Eurydice |
+| `authorities` | Schulbehörden | Regierungspräsidien |
+
+## Tagging-Regeln
+
+Die YAML-Regeldateien im `rules/` Verzeichnis definieren das Tagging:
+
+- `doc_type_rules.yaml` - Dokumenttyp-Erkennung
+- `subject_rules.yaml` - Fächer-Erkennung
+- `level_rules.yaml` - Schulstufen-Erkennung
+- `trust_rules.yaml` - Trust-Score-Berechnung
+
+### Beispiel: doc_type_rules.yaml
+
+```yaml
+doc_types:
+  Lehrplan:
+    strong_terms:
+      - Lehrplan
+      - Kernlehrplan
+      - Bildungsplan
+    medium_terms:
+      - Curriculum
+      - Kompetenzerwartungen
+    url_patterns:
+      - /lehrplan
+      - /kernlehrplan
+
+priority_order:
+  - Pruefung_Abitur
+  - Lehrplan
+  - Arbeitsblatt
+```
+
+## Projektstruktur
+
+```
+edu-search-service/
+├── cmd/
+│   └── server/
+│       └── main.go          # Entry Point
+├── internal/
+│   ├── api/
+│   │   └── handlers/
+│   │       ├── handlers.go       # Search & Health Handler
+│   │       └── admin_handlers.go # Admin API Handler
+│   ├── config/
+│   │   └── config.go        # Konfiguration
+│   ├── crawler/
+│   │   ├── crawler.go       # URL Fetcher
+│   │   └── api_client.go    # Backend API Client (Seeds)
+│   ├── robots/
+│   │   └── robots.go        # robots.txt Parser & Checker
+│   ├── embedding/
+│   │   └── embedding.go     # Embedding Provider (OpenAI/Ollama)
+│   ├── extractor/
+│   │   └── extractor.go     # HTML/PDF Extraktion
+│   ├── indexer/
+│   │   └── mapping.go       # OpenSearch Indexer
+│   ├── pipeline/
+│   │   └── pipeline.go      # Crawl Orchestrierung
+│   ├── quality/
+│   │   └── quality.go       # Multi-Faktor Quality Scoring
+│   ├── scheduler/
+│   │   └── scheduler.go     # Automatisches Crawl-Scheduling
+│   ├── search/
+│   │   └── search.go        # Search Service (Keyword/Semantic/Hybrid)
+│   └── tagger/
+│       └── tagger.go        # Regelbasiertes Tagging
+├── rules/
+│   ├── doc_type_rules.yaml
+│   ├── subject_rules.yaml
+│   ├── level_rules.yaml
+│   └── trust_rules.yaml
+├── seeds/
+│   ├── federal.txt
+│   ├── states.txt
+│   └── denylist.txt
+├── Dockerfile
+├── docker-compose.yml
+├── go.mod
+└── README.md
+```
+
+## Abhängigkeiten
+
+| Package | Version | Beschreibung | Lizenz |
+|---------|---------|--------------|--------|
+| `github.com/gin-gonic/gin` | v1.9+ | HTTP Framework | MIT |
+| `github.com/opensearch-project/opensearch-go/v2` | v2.3+ | OpenSearch Client | Apache-2.0 |
+| `github.com/PuerkitoBio/goquery` | v1.8+ | HTML Parser | BSD-3-Clause |
+| `github.com/ledongthuc/pdf` | v0.0.0-20240201 | PDF Text Extraktion | MIT |
+| `gopkg.in/yaml.v3` | v3.0+ | YAML Parser | MIT |
+| `github.com/google/uuid` | v1.4+ | UUID Generation | BSD-3-Clause |
+| `golang.org/x/net` | v0.19+ | HTML Utilities | BSD-3-Clause |
+
+## Tests ausführen
+
+```bash
+# Alle Tests
+go test -v ./...
+
+# Mit Coverage
+go test -cover ./...
+
+# Nur Tagger Tests
+go test -v ./internal/tagger/...
+
+# Nur Crawler Tests
+go test -v ./internal/crawler/...
+```
+
+## Lizenz
+
+Proprietär - BreakPilot GmbH
+
+## Kontakt
+
+- Security Issues: security@breakpilot.com
+- Bugs: https://github.com/breakpilot/edu-search-service/issues
--- a/edu-search-service/cmd/server/main.go
+++ b/edu-search-service/cmd/server/main.go
@@ -0,0 +1,187 @@
+package main
+
+import (
+	"context"
+	"log"
+	"net/http"
+	"os"
+	"os/signal"
+	"syscall"
+	"time"
+
+	"github.com/breakpilot/edu-search-service/internal/api/handlers"
+	"github.com/breakpilot/edu-search-service/internal/config"
+	"github.com/breakpilot/edu-search-service/internal/database"
+	"github.com/breakpilot/edu-search-service/internal/indexer"
+	"github.com/breakpilot/edu-search-service/internal/orchestrator"
+	"github.com/breakpilot/edu-search-service/internal/search"
+	"github.com/breakpilot/edu-search-service/internal/staff"
+	"github.com/gin-gonic/gin"
+)
+
+func main() {
+	log.Println("Starting edu-search-service...")
+
+	// Load configuration
+	cfg := config.Load()
+	log.Printf("Configuration loaded: Port=%s, OpenSearch=%s, Index=%s",
+		cfg.Port, cfg.OpenSearchURL, cfg.IndexName)
+
+	// Initialize OpenSearch indexer client
+	indexClient, err := indexer.NewClient(
+		cfg.OpenSearchURL,
+		cfg.OpenSearchUsername,
+		cfg.OpenSearchPassword,
+		cfg.IndexName,
+	)
+	if err != nil {
+		log.Fatalf("Failed to create indexer client: %v", err)
+	}
+
+	// Create index if not exists
+	ctx := context.Background()
+	if err := indexClient.CreateIndex(ctx); err != nil {
+		log.Printf("Warning: Could not create index (may already exist): %v", err)
+	}
+
+	// Initialize search service
+	searchService, err := search.NewService(
+		cfg.OpenSearchURL,
+		cfg.OpenSearchUsername,
+		cfg.OpenSearchPassword,
+		cfg.IndexName,
+	)
+	if err != nil {
+		log.Fatalf("Failed to create search service: %v", err)
+	}
+
+	// Initialize seed store for admin API
+	if err := handlers.InitSeedStore(cfg.SeedsDir); err != nil {
+		log.Printf("Warning: Could not initialize seed store: %v", err)
+	}
+
+	// Create handler
+	handler := handlers.NewHandler(cfg, searchService, indexClient)
+
+	// Initialize PostgreSQL for Staff/Publications database
+	dbCfg := &database.Config{
+		Host:     cfg.DBHost,
+		Port:     cfg.DBPort,
+		User:     cfg.DBUser,
+		Password: cfg.DBPassword,
+		DBName:   cfg.DBName,
+		SSLMode:  cfg.DBSSLMode,
+	}
+
+	db, err := database.New(ctx, dbCfg)
+	if err != nil {
+		log.Printf("Warning: Could not connect to PostgreSQL for staff database: %v", err)
+		log.Println("Staff/Publications features will be disabled")
+	} else {
+		defer db.Close()
+		log.Println("Connected to PostgreSQL for staff/publications database")
+
+		// Run migrations
+		if err := db.RunMigrations(ctx); err != nil {
+			log.Printf("Warning: Could not run migrations: %v", err)
+		}
+	}
+
+	// Create repository for Staff handlers (may be nil if DB connection failed)
+	var repo *database.Repository
+	if db != nil {
+		repo = database.NewRepository(db)
+	}
+
+	// Setup Gin router
+	gin.SetMode(gin.ReleaseMode)
+	router := gin.New()
+	router.Use(gin.Recovery())
+	router.Use(gin.Logger())
+
+	// CORS middleware
+	router.Use(func(c *gin.Context) {
+		c.Writer.Header().Set("Access-Control-Allow-Origin", "*")
+		c.Writer.Header().Set("Access-Control-Allow-Methods", "GET, POST, OPTIONS")
+		c.Writer.Header().Set("Access-Control-Allow-Headers", "Content-Type, Authorization")
+
+		if c.Request.Method == "OPTIONS" {
+			c.AbortWithStatus(204)
+			return
+		}
+
+		c.Next()
+	})
+
+	// Setup routes
+	handlers.SetupRoutes(router, handler, cfg.APIKey)
+
+	// Setup Staff/Publications routes if database is available
+	if repo != nil {
+		staffHandlers := handlers.NewStaffHandlers(repo, cfg.StaffCrawlerEmail)
+		apiV1 := router.Group("/api/v1")
+		staffHandlers.RegisterRoutes(apiV1)
+		log.Println("Staff/Publications API routes registered")
+
+		// Setup AI Extraction routes for vast.ai integration
+		aiHandlers := handlers.NewAIExtractionHandlers(repo)
+		aiHandlers.RegisterRoutes(apiV1)
+		log.Println("AI Extraction API routes registered")
+	}
+
+	// Setup Orchestrator routes if database is available
+	if db != nil {
+		orchRepo := orchestrator.NewPostgresRepository(db.Pool)
+
+		// Create real crawlers with adapters for orchestrator interface
+		staffCrawler := staff.NewStaffCrawler(repo)
+		staffAdapter := staff.NewOrchestratorAdapter(staffCrawler, repo)
+		pubAdapter := staff.NewPublicationOrchestratorAdapter(repo)
+
+		orch := orchestrator.NewOrchestrator(orchRepo, staffAdapter, pubAdapter)
+		orchHandler := handlers.NewOrchestratorHandler(orch, orchRepo)
+
+		v1 := router.Group("/v1")
+		v1.Use(handlers.AuthMiddleware(cfg.APIKey))
+		handlers.SetupOrchestratorRoutes(v1, orchHandler)
+		log.Println("Orchestrator API routes registered")
+
+		// Setup Audience routes (reuses orchRepo which implements AudienceRepository)
+		audienceHandler := handlers.NewAudienceHandler(orchRepo)
+		handlers.SetupAudienceRoutes(v1, audienceHandler)
+		log.Println("Audience API routes registered")
+	}
+
+	// Create HTTP server
+	srv := &http.Server{
+		Addr:         ":" + cfg.Port,
+		Handler:      router,
+		ReadTimeout:  10 * time.Second,
+		WriteTimeout: 30 * time.Second,
+		IdleTimeout:  60 * time.Second,
+	}
+
+	// Start server in goroutine
+	go func() {
+		log.Printf("Server listening on port %s", cfg.Port)
+		if err := srv.ListenAndServe(); err != nil && err != http.ErrServerClosed {
+			log.Fatalf("Server error: %v", err)
+		}
+	}()
+
+	// Graceful shutdown
+	quit := make(chan os.Signal, 1)
+	signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
+	<-quit
+
+	log.Println("Shutting down server...")
+
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+
+	if err := srv.Shutdown(ctx); err != nil {
+		log.Fatalf("Server forced to shutdown: %v", err)
+	}
+
+	log.Println("Server exited")
+}
--- a/edu-search-service/go.mod
+++ b/edu-search-service/go.mod
@@ -0,0 +1,46 @@
+module github.com/breakpilot/edu-search-service
+
+go 1.23
+
+require (
+	github.com/PuerkitoBio/goquery v1.8.1
+	github.com/gin-gonic/gin v1.9.1
+	github.com/google/uuid v1.4.0
+	github.com/jackc/pgx/v5 v5.5.1
+	github.com/ledongthuc/pdf v0.0.0-20240201131950-da5b75280b06
+	github.com/opensearch-project/opensearch-go/v2 v2.3.0
+	golang.org/x/net v0.19.0
+	gopkg.in/yaml.v3 v3.0.1
+)
+
+require (
+	github.com/andybalholm/cascadia v1.3.1 // indirect
+	github.com/bytedance/sonic v1.9.1 // indirect
+	github.com/chenzhuoyu/base64x v0.0.0-20221115062448-fe3a3abad311 // indirect
+	github.com/gabriel-vasile/mimetype v1.4.2 // indirect
+	github.com/gin-contrib/sse v0.1.0 // indirect
+	github.com/go-playground/locales v0.14.1 // indirect
+	github.com/go-playground/universal-translator v0.18.1 // indirect
+	github.com/go-playground/validator/v10 v10.14.0 // indirect
+	github.com/goccy/go-json v0.10.2 // indirect
+	github.com/jackc/pgpassfile v1.0.0 // indirect
+	github.com/jackc/pgservicefile v0.0.0-20221227161230-091c0ba34f0a // indirect
+	github.com/jackc/puddle/v2 v2.2.1 // indirect
+	github.com/json-iterator/go v1.1.12 // indirect
+	github.com/klauspost/cpuid/v2 v2.2.4 // indirect
+	github.com/kr/text v0.2.0 // indirect
+	github.com/leodido/go-urn v1.2.4 // indirect
+	github.com/mattn/go-isatty v0.0.19 // indirect
+	github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd // indirect
+	github.com/modern-go/reflect2 v1.0.2 // indirect
+	github.com/pelletier/go-toml/v2 v2.0.8 // indirect
+	github.com/rogpeppe/go-internal v1.14.1 // indirect
+	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
+	github.com/ugorji/go/codec v1.2.11 // indirect
+	golang.org/x/arch v0.3.0 // indirect
+	golang.org/x/crypto v0.16.0 // indirect
+	golang.org/x/sync v0.1.0 // indirect
+	golang.org/x/sys v0.26.0 // indirect
+	golang.org/x/text v0.14.0 // indirect
+	google.golang.org/protobuf v1.30.0 // indirect
+)
--- a/edu-search-service/go.sum
+++ b/edu-search-service/go.sum
@@ -0,0 +1,165 @@
+github.com/PuerkitoBio/goquery v1.8.1 h1:uQxhNlArOIdbrH1tr0UXwdVFgDcZDrZVdcpygAcwmWM=
+github.com/PuerkitoBio/goquery v1.8.1/go.mod h1:Q8ICL1kNUJ2sXGoAhPGUdYDJvgQgHzJsnnd3H7Ho5jQ=
+github.com/andybalholm/cascadia v1.3.1 h1:nhxRkql1kdYCc8Snf7D5/D3spOX+dBgjA6u8x004T2c=
+github.com/andybalholm/cascadia v1.3.1/go.mod h1:R4bJ1UQfqADjvDa4P6HZHLh/3OxWWEqc0Sk8XGwHqvA=
+github.com/aws/aws-sdk-go v1.44.263/go.mod h1:aVsgQcEevwlmQ7qHE9I3h+dtQgpqhFB+i8Phjh7fkwI=
+github.com/aws/aws-sdk-go-v2 v1.18.0/go.mod h1:uzbQtefpm44goOPmdKyAlXSNcwlRgF3ePWVW6EtJvvw=
+github.com/aws/aws-sdk-go-v2/config v1.18.25/go.mod h1:dZnYpD5wTW/dQF0rRNLVypB396zWCcPiBIvdvSWHEg4=
+github.com/aws/aws-sdk-go-v2/credentials v1.13.24/go.mod h1:jYPYi99wUOPIFi0rhiOvXeSEReVOzBqFNOX5bXYoG2o=
+github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.13.3/go.mod h1:4Q0UFP0YJf0NrsEuEYHpM9fTSEVnD16Z3uyEF7J9JGM=
+github.com/aws/aws-sdk-go-v2/internal/configsources v1.1.33/go.mod h1:7i0PF1ME/2eUPFcjkVIwq+DOygHEoK92t5cDqNgYbIw=
+github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.4.27/go.mod h1:UrHnn3QV/d0pBZ6QBAEQcqFLf8FAzLmoUfPVIueOvoM=
+github.com/aws/aws-sdk-go-v2/internal/ini v1.3.34/go.mod h1:Etz2dj6UHYuw+Xw830KfzCfWGMzqvUTCjUj5b76GVDc=
+github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.9.27/go.mod h1:EOwBD4J4S5qYszS5/3DpkejfuK+Z5/1uzICfPaZLtqw=
+github.com/aws/aws-sdk-go-v2/service/sso v1.12.10/go.mod h1:ouy2P4z6sJN70fR3ka3wD3Ro3KezSxU6eKGQI2+2fjI=
+github.com/aws/aws-sdk-go-v2/service/ssooidc v1.14.10/go.mod h1:AFvkxc8xfBe8XA+5St5XIHHrQQtkxqrRincx4hmMHOk=
+github.com/aws/aws-sdk-go-v2/service/sts v1.19.0/go.mod h1:BgQOMsg8av8jset59jelyPW7NoZcZXLVpDsXunGDrk8=
+github.com/aws/smithy-go v1.13.5/go.mod h1:Tg+OJXh4MB2R/uN61Ko2f6hTZwB/ZYGOtib8J3gBHzA=
+github.com/bytedance/sonic v1.5.0/go.mod h1:ED5hyg4y6t3/9Ku1R6dU/4KyJ48DZ4jPhfY1O2AihPM=
+github.com/bytedance/sonic v1.9.1 h1:6iJ6NqdoxCDr6mbY8h18oSO+cShGSMRGCEo7F2h0x8s=
+github.com/bytedance/sonic v1.9.1/go.mod h1:i736AoUSYt75HyZLoJW9ERYxcy6eaN6h4BZXU064P/U=
+github.com/chenzhuoyu/base64x v0.0.0-20211019084208-fb5309c8db06/go.mod h1:DH46F32mSOjUmXrMHnKwZdA8wcEefY7UVqBKYGjpdQY=
+github.com/chenzhuoyu/base64x v0.0.0-20221115062448-fe3a3abad311 h1:qSGYFH7+jGhDF8vLC+iwCD4WpbV1EBDSzWkJODFLams=
+github.com/chenzhuoyu/base64x v0.0.0-20221115062448-fe3a3abad311/go.mod h1:b583jCggY9gE99b6G5LEC39OIiVsWj+R97kbl5odCEk=
+github.com/creack/pty v1.1.9/go.mod h1:oKZEueFk5CKHvIhNR5MUki03XCEU+Q6VDXinZuGJ33E=
+github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
+github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/gabriel-vasile/mimetype v1.4.2 h1:w5qFW6JKBz9Y393Y4q372O9A7cUSequkh1Q7OhCmWKU=
+github.com/gabriel-vasile/mimetype v1.4.2/go.mod h1:zApsH/mKG4w07erKIaJPFiX0Tsq9BFQgN3qGY5GnNgA=
+github.com/gin-contrib/sse v0.1.0 h1:Y/yl/+YNO8GZSjAhjMsSuLt29uWRFHdHYUb5lYOV9qE=
+github.com/gin-contrib/sse v0.1.0/go.mod h1:RHrZQHXnP2xjPF+u1gW/2HnVO7nvIa9PG3Gm+fLHvGI=
+github.com/gin-gonic/gin v1.9.1 h1:4idEAncQnU5cB7BeOkPtxjfCSye0AAm1R0RVIqJ+Jmg=
+github.com/gin-gonic/gin v1.9.1/go.mod h1:hPrL7YrpYKXt5YId3A/Tnip5kqbEAP+KLuI3SUcPTeU=
+github.com/go-playground/assert/v2 v2.2.0 h1:JvknZsQTYeFEAhQwI4qEt9cyV5ONwRHC+lYKSsYSR8s=
+github.com/go-playground/assert/v2 v2.2.0/go.mod h1:VDjEfimB/XKnb+ZQfWdccd7VUvScMdVu0Titje2rxJ4=
+github.com/go-playground/locales v0.14.1 h1:EWaQ/wswjilfKLTECiXz7Rh+3BjFhfDFKv/oXslEjJA=
+github.com/go-playground/locales v0.14.1/go.mod h1:hxrqLVvrK65+Rwrd5Fc6F2O76J/NuW9t0sjnWqG1slY=
+github.com/go-playground/universal-translator v0.18.1 h1:Bcnm0ZwsGyWbCzImXv+pAJnYK9S473LQFuzCbDbfSFY=
+github.com/go-playground/universal-translator v0.18.1/go.mod h1:xekY+UJKNuX9WP91TpwSH2VMlDf28Uj24BCp08ZFTUY=
+github.com/go-playground/validator/v10 v10.14.0 h1:vgvQWe3XCz3gIeFDm/HnTIbj6UGmg/+t63MyGU2n5js=
+github.com/go-playground/validator/v10 v10.14.0/go.mod h1:9iXMNT7sEkjXb0I+enO7QXmzG6QCsPWY4zveKFVRSyU=
+github.com/goccy/go-json v0.10.2 h1:CrxCmQqYDkv1z7lO7Wbh2HN93uovUHgrECaO5ZrCXAU=
+github.com/goccy/go-json v0.10.2/go.mod h1:6MelG93GURQebXPDq3khkgXZkazVtN9CRI+MGFi0w8I=
+github.com/golang/protobuf v1.5.0/go.mod h1:FsONVRAS9T7sI+LIUmWTfcYkHO4aIWwzhcaSAoJOfIk=
+github.com/google/go-cmp v0.5.5/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
+github.com/google/go-cmp v0.5.8 h1:e6P7q2lk1O+qJJb4BtCQXlK8vWEO8V1ZeuEdJNOqZyg=
+github.com/google/go-cmp v0.5.8/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
+github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
+github.com/google/uuid v1.4.0 h1:MtMxsa51/r9yyhkyLsVeVt0B+BGQZzpQiTQ4eHZ8bc4=
+github.com/google/uuid v1.4.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/jackc/pgpassfile v1.0.0 h1:/6Hmqy13Ss2zCq62VdNG8tM1wchn8zjSGOBJ6icpsIM=
+github.com/jackc/pgpassfile v1.0.0/go.mod h1:CEx0iS5ambNFdcRtxPj5JhEz+xB6uRky5eyVu/W2HEg=
+github.com/jackc/pgservicefile v0.0.0-20221227161230-091c0ba34f0a h1:bbPeKD0xmW/Y25WS6cokEszi5g+S0QxI/d45PkRi7Nk=
+github.com/jackc/pgservicefile v0.0.0-20221227161230-091c0ba34f0a/go.mod h1:5TJZWKEWniPve33vlWYSoGYefn3gLQRzjfDlhSJ9ZKM=
+github.com/jackc/pgx/v5 v5.5.1 h1:5I9etrGkLrN+2XPCsi6XLlV5DITbSL/xBZdmAxFcXPI=
+github.com/jackc/pgx/v5 v5.5.1/go.mod h1:Ig06C2Vu0t5qXC60W8sqIthScaEnFvojjj9dSljmHRA=
+github.com/jackc/puddle/v2 v2.2.1 h1:RhxXJtFG022u4ibrCSMSiu5aOq1i77R3OHKNJj77OAk=
+github.com/jackc/puddle/v2 v2.2.1/go.mod h1:vriiEXHvEE654aYKXXjOvZM39qJ0q+azkZFrfEOc3H4=
+github.com/jmespath/go-jmespath v0.4.0/go.mod h1:T8mJZnbsbmF+m6zOOFylbeCJqk5+pHWvzYPziyZiYoo=
+github.com/jmespath/go-jmespath/internal/testify v1.5.1/go.mod h1:L3OGu8Wl2/fWfCI6z80xFu9LTZmf1ZRjMHUOPmWr69U=
+github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnrnM=
+github.com/json-iterator/go v1.1.12/go.mod h1:e30LSqwooZae/UwlEbR2852Gd8hjQvJoHmT4TnhNGBo=
+github.com/klauspost/cpuid/v2 v2.0.9/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
+github.com/klauspost/cpuid/v2 v2.2.4 h1:acbojRNwl3o09bUq+yDCtZFc1aiwaAAxtcn8YkZXnvk=
+github.com/klauspost/cpuid/v2 v2.2.4/go.mod h1:RVVoqg1df56z8g3pUjL/3lE5UfnlrJX8tyFgg4nqhuY=
+github.com/kr/pretty v0.3.0 h1:WgNl7dwNpEZ6jJ9k1snq4pZsg7DOEN8hP9Xw0Tsjwk0=
+github.com/kr/pretty v0.3.0/go.mod h1:640gp4NfQd8pI5XOwp5fnNeVWj67G7CFk/SaSQn7NBk=
+github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
+github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
+github.com/ledongthuc/pdf v0.0.0-20240201131950-da5b75280b06 h1:kacRlPN7EN++tVpGUorNGPn/4DnB7/DfTY82AOn6ccU=
+github.com/ledongthuc/pdf v0.0.0-20240201131950-da5b75280b06/go.mod h1:imJHygn/1yfhB7XSJJKlFZKl/J+dCPAknuiaGOshXAs=
+github.com/leodido/go-urn v1.2.4 h1:XlAE/cm/ms7TE/VMVoduSpNBoyc2dOxHs5MZSwAN63Q=
+github.com/leodido/go-urn v1.2.4/go.mod h1:7ZrI8mTSeBSHl/UaRyKQW1qZeMgak41ANeCNaVckg+4=
+github.com/mattn/go-isatty v0.0.19 h1:JITubQf0MOLdlGRuRq+jtsDlekdYPia9ZFsB8h/APPA=
+github.com/mattn/go-isatty v0.0.19/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/modern-go/concurrent v0.0.0-20180228061459-e0a39a4cb421/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
+github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd h1:TRLaZ9cD/w8PVh93nsPXa1VrQ6jlwL5oN8l14QlcNfg=
+github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
+github.com/modern-go/reflect2 v1.0.2 h1:xBagoLtFs94CBntxluKeaWgTMpvLxC4ur3nMaC9Gz0M=
+github.com/modern-go/reflect2 v1.0.2/go.mod h1:yWuevngMOJpCy52FWWMvUC8ws7m/LJsjYzDa0/r8luk=
+github.com/opensearch-project/opensearch-go/v2 v2.3.0 h1:nQIEMr+A92CkhHrZgUhcfsrZjibvB3APXf2a1VwCmMQ=
+github.com/opensearch-project/opensearch-go/v2 v2.3.0/go.mod h1:8LDr9FCgUTVoT+5ESjc2+iaZuldqE+23Iq0r1XeNue8=
+github.com/pelletier/go-toml/v2 v2.0.8 h1:0ctb6s9mE31h0/lhu+J6OPmVeDxJn+kYnJc2jZR9tGQ=
+github.com/pelletier/go-toml/v2 v2.0.8/go.mod h1:vuYfssBdrU2XDZ9bYydBu6t+6a6PYNcZljzZR9VXg+4=
+github.com/pkg/errors v0.9.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
+github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
+github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
+github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
+github.com/rogpeppe/go-internal v1.14.1/go.mod h1:MaRKkUm5W0goXpeCfT7UZI6fk/L7L7so1lCWt35ZSgc=
+github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
+github.com/stretchr/objx v0.4.0/go.mod h1:YvHI0jy2hoMjB+UWwv71VJQ9isScKT/TqJzVSSt89Yw=
+github.com/stretchr/objx v0.5.0/go.mod h1:Yh+to48EsGEfYuaHDzXPcE3xhTkx73EhmCGUpEOglKo=
+github.com/stretchr/testify v1.3.0/go.mod h1:M5WIy9Dh21IEIfnGCwXGc5bZfKNJtfHm1UVUgZn+9EI=
+github.com/stretchr/testify v1.7.0/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
+github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
+github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
+github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
+github.com/stretchr/testify v1.8.2/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
+github.com/stretchr/testify v1.8.3 h1:RP3t2pwF7cMEbC1dqtB6poj3niw/9gnV4Cjg5oW5gtY=
+github.com/stretchr/testify v1.8.3/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
+github.com/twitchyliquid64/golang-asm v0.15.1 h1:SU5vSMR7hnwNxj24w34ZyCi/FmDZTkS4MhqMhdFk5YI=
+github.com/twitchyliquid64/golang-asm v0.15.1/go.mod h1:a1lVb/DtPvCB8fslRZhAngC2+aY1QWCk3Cedj/Gdt08=
+github.com/ugorji/go/codec v1.2.11 h1:BMaWp1Bb6fHwEtbplGBGJ498wD+LKlNSl25MjdZY4dU=
+github.com/ugorji/go/codec v1.2.11/go.mod h1:UNopzCgEMSXjBc6AOMqYvWC1ktqTAfzJZUZgYf6w6lg=
+github.com/yuin/goldmark v1.4.13/go.mod h1:6yULJ656Px+3vBD8DxQVa3kxgyrAnzto9xy5taEt/CY=
+golang.org/x/arch v0.0.0-20210923205945-b76863e36670/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
+golang.org/x/arch v0.3.0 h1:02VY4/ZcO/gBOH6PUaoiptASxtXU10jazRCP865E97k=
+golang.org/x/arch v0.3.0/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
+golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
+golang.org/x/crypto v0.0.0-20210921155107-089bfa567519/go.mod h1:GvvjBRRGRdwPK5ydBHafDWAxML/pGHZbMvKqRZ5+Abc=
+golang.org/x/crypto v0.16.0 h1:mMMrFzRSCF0GvB7Ne27XVtVAaXLrPmgPC7/v0tkwHaY=
+golang.org/x/crypto v0.16.0/go.mod h1:gCAAfMLgwOJRpTjQ2zCCt2OcSfYMTeZVSRtQlPC7Nq4=
+golang.org/x/mod v0.6.0-dev.0.20220419223038-86c51ed26bb4/go.mod h1:jJ57K6gSWd91VN4djpZkiMVwK6gcyfeH4XE8wZrZaV4=
+golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20210226172049-e18ecbb05110/go.mod h1:m0MpNAwzfU5UDzcl9v0D8zg8gWTRqZa9RBIspLL5mdg=
+golang.org/x/net v0.0.0-20210916014120-12bc252f5db8/go.mod h1:9nx3DQGgdP8bBQD5qxJ1jj9UTztislL4KSBs9R2vV5Y=
+golang.org/x/net v0.0.0-20220722155237-a158d28d115b/go.mod h1:XRhObCWvk6IyKnWLug+ECip1KBveYUHfp+8e9klMJ9c=
+golang.org/x/net v0.1.0/go.mod h1:Cx3nUiGt4eDBEyega/BKRp+/AlGL8hYe7U9odMt2Cco=
+golang.org/x/net v0.7.0/go.mod h1:2Tu9+aMcznHK/AK1HMvgo6xiTLG5rD5rZLDS+rp2Bjs=
+golang.org/x/net v0.19.0 h1:zTwKpTd2XuCqf8huc7Fo2iSy+4RHPd10s4KzeTnVr1c=
+golang.org/x/net v0.19.0/go.mod h1:CfAk/cbD4CthTvqiEl8NpboMuiuOYsAr/7NOjZJtv1U=
+golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20220722155255-886fb9371eb4/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.1.0 h1:wsuoTGHzEhffawBOhz5CYhcrV4IdKZbEyZjBMuTp12o=
+golang.org/x/sync v0.1.0/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
+golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210423082822-04245dca01da/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210615035016-665e8c7367d1/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.0.0-20220520151302-bc2c85ada10a/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.0.0-20220704084225-05e143d24a9e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.0.0-20220722155257-8c9f86f7a55f/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.1.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.5.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.26.0 h1:KHjCJyddX0LoSTb3J+vWpupP9p0oznkqVk/IfjymZbo=
+golang.org/x/sys v0.26.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
+golang.org/x/term v0.0.0-20210927222741-03fcf44c2211/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
+golang.org/x/term v0.1.0/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
+golang.org/x/term v0.5.0/go.mod h1:jMB1sMXY+tzblOD4FWmEbocvup2/aLOaQEp7JmGp78k=
+golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
+golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
+golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
+golang.org/x/text v0.3.7/go.mod h1:u+2+/6zg+i71rQMx5EYifcz6MCKuco9NR6JIITiCfzQ=
+golang.org/x/text v0.4.0/go.mod h1:mrYo+phRRbMaCq/xk9113O4dZlRixOauAjOtrjsXDZ8=
+golang.org/x/text v0.7.0/go.mod h1:mrYo+phRRbMaCq/xk9113O4dZlRixOauAjOtrjsXDZ8=
+golang.org/x/text v0.14.0 h1:ScX5w1eTa3QqT8oi6+ziP7dTV1S2+ALU0bI+0zXKWiQ=
+golang.org/x/text v0.14.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
+golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.1.12/go.mod h1:hNGJHUnrk76NpqgfD5Aqm5Crs+Hm0VOH/i9J2+nxYbc=
+golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+google.golang.org/protobuf v1.26.0-rc.1/go.mod h1:jlhhOSvTdKEhbULTjvd4ARK9grFBp09yW+WbY/TyQbw=
+google.golang.org/protobuf v1.30.0 h1:kPPoIgf3TsEvrm0PFe15JQ+570QVxYzEvvHqChK+cng=
+google.golang.org/protobuf v1.30.0/go.mod h1:HV8QOd/L58Z+nl8r43ehVNZIU/HEI6OcFqwMG9pJV4I=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c h1:Hei/4ADfdWqJk1ZMxUNpqntNwaWcugrBjAiHlqqRiVk=
+gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c/go.mod h1:JHkPIbrfpd72SG/EVd6muEfDQjcINNoR0C8j2r3qZ4Q=
+gopkg.in/yaml.v2 v2.2.8/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
+gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
+gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+rsc.io/pdf v0.1.1/go.mod h1:n8OzWcQ6Sp37PL01nO98y4iUCRdTGarVfzxY20ICaU4=
--- a/edu-search-service/internal/api/handlers/admin_handlers.go
+++ b/edu-search-service/internal/api/handlers/admin_handlers.go
@@ -0,0 +1,406 @@
+package handlers
+
+import (
+	"encoding/json"
+	"net/http"
+	"os"
+	"path/filepath"
+	"sync"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+)
+
+// SeedURL represents a seed URL configuration
+type SeedURL struct {
+	ID            string    `json:"id"`
+	URL           string    `json:"url"`
+	Category      string    `json:"category"`
+	Name          string    `json:"name"`
+	Description   string    `json:"description"`
+	TrustBoost    float64   `json:"trustBoost"`
+	Enabled       bool      `json:"enabled"`
+	LastCrawled   *string   `json:"lastCrawled,omitempty"`
+	DocumentCount int       `json:"documentCount,omitempty"`
+	CreatedAt     time.Time `json:"createdAt"`
+	UpdatedAt     time.Time `json:"updatedAt"`
+}
+
+// CrawlStats contains crawl statistics
+type CrawlStats struct {
+	TotalDocuments        int                `json:"totalDocuments"`
+	TotalSeeds            int                `json:"totalSeeds"`
+	LastCrawlTime         *string            `json:"lastCrawlTime,omitempty"`
+	CrawlStatus           string             `json:"crawlStatus"`
+	DocumentsPerCategory  map[string]int     `json:"documentsPerCategory"`
+	DocumentsPerDocType   map[string]int     `json:"documentsPerDocType"`
+	AvgTrustScore         float64            `json:"avgTrustScore"`
+}
+
+// SeedStore manages seed URLs in memory and file
+type SeedStore struct {
+	seeds    map[string]SeedURL
+	mu       sync.RWMutex
+	filePath string
+}
+
+var seedStore *SeedStore
+var crawlStatus = "idle"
+var lastCrawlTime *string
+
+// InitSeedStore initializes the seed store
+func InitSeedStore(seedsDir string) error {
+	seedStore = &SeedStore{
+		seeds:    make(map[string]SeedURL),
+		filePath: filepath.Join(seedsDir, "seeds.json"),
+	}
+
+	// Try to load existing seeds from JSON file
+	if err := seedStore.loadFromFile(); err != nil {
+		// If file doesn't exist, load from txt files
+		return seedStore.loadFromTxtFiles(seedsDir)
+	}
+	return nil
+}
+
+func (s *SeedStore) loadFromFile() error {
+	data, err := os.ReadFile(s.filePath)
+	if err != nil {
+		return err
+	}
+
+	var seeds []SeedURL
+	if err := json.Unmarshal(data, &seeds); err != nil {
+		return err
+	}
+
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	for _, seed := range seeds {
+		s.seeds[seed.ID] = seed
+	}
+	return nil
+}
+
+func (s *SeedStore) loadFromTxtFiles(seedsDir string) error {
+	// Default seeds from category files
+	defaultSeeds := []SeedURL{
+		{ID: uuid.New().String(), URL: "https://www.kmk.org", Category: "federal", Name: "Kultusministerkonferenz", Description: "Beschlüsse und Bildungsstandards", TrustBoost: 0.50, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.bildungsserver.de", Category: "federal", Name: "Deutscher Bildungsserver", Description: "Zentrale Bildungsinformationen", TrustBoost: 0.50, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.bpb.de", Category: "federal", Name: "Bundeszentrale politische Bildung", Description: "Politische Bildung", TrustBoost: 0.45, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.bmbf.de", Category: "federal", Name: "BMBF", Description: "Bundesbildungsministerium", TrustBoost: 0.50, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.iqb.hu-berlin.de", Category: "federal", Name: "IQB", Description: "Institut Qualitätsentwicklung", TrustBoost: 0.50, Enabled: true},
+
+		// Science
+		{ID: uuid.New().String(), URL: "https://www.bertelsmann-stiftung.de/de/themen/bildung", Category: "science", Name: "Bertelsmann Stiftung", Description: "Bildungsstudien und Ländermonitor", TrustBoost: 0.40, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.oecd.org/pisa", Category: "science", Name: "PISA-Studien", Description: "Internationale Schulleistungsstudie", TrustBoost: 0.45, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.iea.nl/studies/iea/pirls", Category: "science", Name: "IGLU/PIRLS", Description: "Internationale Grundschul-Lese-Untersuchung", TrustBoost: 0.45, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.iea.nl/studies/iea/timss", Category: "science", Name: "TIMSS", Description: "Trends in International Mathematics and Science Study", TrustBoost: 0.45, Enabled: true},
+
+		// Bundesländer
+		{ID: uuid.New().String(), URL: "https://www.km.bayern.de", Category: "states", Name: "Bayern Kultusministerium", Description: "Lehrpläne Bayern", TrustBoost: 0.45, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.schulministerium.nrw", Category: "states", Name: "NRW Schulministerium", Description: "Lehrpläne NRW", TrustBoost: 0.45, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.berlin.de/sen/bildung", Category: "states", Name: "Berlin Bildung", Description: "Rahmenlehrpläne Berlin", TrustBoost: 0.45, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://kultusministerium.hessen.de", Category: "states", Name: "Hessen Kultusministerium", Description: "Kerncurricula Hessen", TrustBoost: 0.45, Enabled: true},
+
+		// Portale
+		{ID: uuid.New().String(), URL: "https://www.lehrer-online.de", Category: "portals", Name: "Lehrer-Online", Description: "Unterrichtsmaterialien", TrustBoost: 0.20, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.4teachers.de", Category: "portals", Name: "4teachers", Description: "Lehrercommunity", TrustBoost: 0.20, Enabled: true},
+		{ID: uuid.New().String(), URL: "https://www.zum.de", Category: "portals", Name: "ZUM", Description: "Zentrale für Unterrichtsmedien", TrustBoost: 0.25, Enabled: true},
+	}
+
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	now := time.Now()
+	for _, seed := range defaultSeeds {
+		seed.CreatedAt = now
+		seed.UpdatedAt = now
+		s.seeds[seed.ID] = seed
+	}
+
+	return s.saveToFile()
+}
+
+func (s *SeedStore) saveToFile() error {
+	seeds := make([]SeedURL, 0, len(s.seeds))
+	for _, seed := range s.seeds {
+		seeds = append(seeds, seed)
+	}
+
+	data, err := json.MarshalIndent(seeds, "", "  ")
+	if err != nil {
+		return err
+	}
+
+	return os.WriteFile(s.filePath, data, 0644)
+}
+
+// GetAllSeeds returns all seeds
+func (s *SeedStore) GetAllSeeds() []SeedURL {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	seeds := make([]SeedURL, 0, len(s.seeds))
+	for _, seed := range s.seeds {
+		seeds = append(seeds, seed)
+	}
+	return seeds
+}
+
+// GetSeed returns a single seed by ID
+func (s *SeedStore) GetSeed(id string) (SeedURL, bool) {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	seed, ok := s.seeds[id]
+	return seed, ok
+}
+
+// CreateSeed adds a new seed
+func (s *SeedStore) CreateSeed(seed SeedURL) (SeedURL, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	seed.ID = uuid.New().String()
+	seed.CreatedAt = time.Now()
+	seed.UpdatedAt = time.Now()
+	s.seeds[seed.ID] = seed
+
+	if err := s.saveToFile(); err != nil {
+		delete(s.seeds, seed.ID)
+		return SeedURL{}, err
+	}
+
+	return seed, nil
+}
+
+// UpdateSeed updates an existing seed
+func (s *SeedStore) UpdateSeed(id string, updates SeedURL) (SeedURL, bool, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	seed, ok := s.seeds[id]
+	if !ok {
+		return SeedURL{}, false, nil
+	}
+
+	// Update fields
+	if updates.URL != "" {
+		seed.URL = updates.URL
+	}
+	if updates.Name != "" {
+		seed.Name = updates.Name
+	}
+	if updates.Category != "" {
+		seed.Category = updates.Category
+	}
+	if updates.Description != "" {
+		seed.Description = updates.Description
+	}
+	seed.TrustBoost = updates.TrustBoost
+	seed.Enabled = updates.Enabled
+	seed.UpdatedAt = time.Now()
+
+	s.seeds[id] = seed
+
+	if err := s.saveToFile(); err != nil {
+		return SeedURL{}, true, err
+	}
+
+	return seed, true, nil
+}
+
+// DeleteSeed removes a seed
+func (s *SeedStore) DeleteSeed(id string) bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if _, ok := s.seeds[id]; !ok {
+		return false
+	}
+
+	delete(s.seeds, id)
+	s.saveToFile()
+	return true
+}
+
+// Admin Handlers
+
+// GetSeeds returns all seed URLs
+func (h *Handler) GetSeeds(c *gin.Context) {
+	if seedStore == nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Seed store not initialized"})
+		return
+	}
+
+	seeds := seedStore.GetAllSeeds()
+	c.JSON(http.StatusOK, seeds)
+}
+
+// CreateSeed adds a new seed URL
+func (h *Handler) CreateSeed(c *gin.Context) {
+	if seedStore == nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Seed store not initialized"})
+		return
+	}
+
+	var seed SeedURL
+	if err := c.ShouldBindJSON(&seed); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	if seed.URL == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "URL is required"})
+		return
+	}
+
+	created, err := seedStore.CreateSeed(seed)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to create seed", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusCreated, created)
+}
+
+// UpdateSeed updates an existing seed URL
+func (h *Handler) UpdateSeed(c *gin.Context) {
+	if seedStore == nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Seed store not initialized"})
+		return
+	}
+
+	id := c.Param("id")
+	if id == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Seed ID required"})
+		return
+	}
+
+	var updates SeedURL
+	if err := c.ShouldBindJSON(&updates); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	updated, found, err := seedStore.UpdateSeed(id, updates)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to update seed", "details": err.Error()})
+		return
+	}
+	if !found {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Seed not found"})
+		return
+	}
+
+	c.JSON(http.StatusOK, updated)
+}
+
+// DeleteSeed removes a seed URL
+func (h *Handler) DeleteSeed(c *gin.Context) {
+	if seedStore == nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Seed store not initialized"})
+		return
+	}
+
+	id := c.Param("id")
+	if id == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Seed ID required"})
+		return
+	}
+
+	if !seedStore.DeleteSeed(id) {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Seed not found"})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{"deleted": true, "id": id})
+}
+
+// GetStats returns crawl statistics
+func (h *Handler) GetStats(c *gin.Context) {
+	// Get document count from OpenSearch
+	totalDocs := 0
+	// TODO: Get real count from OpenSearch
+
+	seeds := []SeedURL{}
+	if seedStore != nil {
+		seeds = seedStore.GetAllSeeds()
+	}
+
+	enabledSeeds := 0
+	for _, seed := range seeds {
+		if seed.Enabled {
+			enabledSeeds++
+		}
+	}
+
+	stats := CrawlStats{
+		TotalDocuments: totalDocs,
+		TotalSeeds:     enabledSeeds,
+		LastCrawlTime:  lastCrawlTime,
+		CrawlStatus:    crawlStatus,
+		DocumentsPerCategory: map[string]int{
+			"federal":      0,
+			"states":       0,
+			"science":      0,
+			"universities": 0,
+			"portals":      0,
+		},
+		DocumentsPerDocType: map[string]int{
+			"Lehrplan":          0,
+			"Arbeitsblatt":      0,
+			"Unterrichtsentwurf": 0,
+			"Erlass_Verordnung": 0,
+			"Pruefung_Abitur":   0,
+			"Studie_Bericht":    0,
+			"Sonstiges":         0,
+		},
+		AvgTrustScore: 0.0,
+	}
+
+	c.JSON(http.StatusOK, stats)
+}
+
+// StartCrawl initiates a crawl run
+func (h *Handler) StartCrawl(c *gin.Context) {
+	if crawlStatus == "running" {
+		c.JSON(http.StatusConflict, gin.H{"error": "Crawl already running"})
+		return
+	}
+
+	crawlStatus = "running"
+
+	// TODO: Start actual crawl in background goroutine
+	go func() {
+		time.Sleep(5 * time.Second) // Simulate crawl
+		now := time.Now().Format(time.RFC3339)
+		lastCrawlTime = &now
+		crawlStatus = "idle"
+	}()
+
+	c.JSON(http.StatusAccepted, gin.H{
+		"status":  "started",
+		"message": "Crawl initiated",
+	})
+}
+
+// SetupAdminRoutes configures admin API routes
+func SetupAdminRoutes(r *gin.RouterGroup, h *Handler) {
+	admin := r.Group("/admin")
+	{
+		// Seeds CRUD
+		admin.GET("/seeds", h.GetSeeds)
+		admin.POST("/seeds", h.CreateSeed)
+		admin.PUT("/seeds/:id", h.UpdateSeed)
+		admin.DELETE("/seeds/:id", h.DeleteSeed)
+
+		// Stats
+		admin.GET("/stats", h.GetStats)
+
+		// Crawl control
+		admin.POST("/crawl/start", h.StartCrawl)
+	}
+}
--- a/edu-search-service/internal/api/handlers/ai_extraction_handlers.go
+++ b/edu-search-service/internal/api/handlers/ai_extraction_handlers.go
@@ -0,0 +1,554 @@
+package handlers
+
+import (
+	"net/http"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+)
+
+// AIExtractionHandlers handles AI-based profile extraction endpoints
+// These endpoints are designed for vast.ai or similar AI services to:
+// 1. Get profile URLs that need extraction
+// 2. Submit extracted data back
+type AIExtractionHandlers struct {
+	repo *database.Repository
+}
+
+// NewAIExtractionHandlers creates new AI extraction handlers
+func NewAIExtractionHandlers(repo *database.Repository) *AIExtractionHandlers {
+	return &AIExtractionHandlers{repo: repo}
+}
+
+// ProfileExtractionTask represents a profile URL to be processed by AI
+type ProfileExtractionTask struct {
+	StaffID       uuid.UUID `json:"staff_id"`
+	ProfileURL    string    `json:"profile_url"`
+	UniversityID  uuid.UUID `json:"university_id"`
+	UniversityURL string    `json:"university_url,omitempty"`
+	FullName      string    `json:"full_name,omitempty"`
+	CurrentData   struct {
+		Email      string `json:"email,omitempty"`
+		Phone      string `json:"phone,omitempty"`
+		Office     string `json:"office,omitempty"`
+		Position   string `json:"position,omitempty"`
+		Department string `json:"department,omitempty"`
+	} `json:"current_data"`
+}
+
+// GetPendingProfiles returns staff profiles that need AI extraction
+// GET /api/v1/ai/extraction/pending?limit=10&university_id=...
+func (h *AIExtractionHandlers) GetPendingProfiles(c *gin.Context) {
+	limit := parseIntDefault(c.Query("limit"), 10)
+	if limit > 100 {
+		limit = 100
+	}
+
+	var universityID *uuid.UUID
+	if uniIDStr := c.Query("university_id"); uniIDStr != "" {
+		id, err := uuid.Parse(uniIDStr)
+		if err == nil {
+			universityID = &id
+		}
+	}
+
+	// Get staff that have profile URLs but missing key data
+	params := database.StaffSearchParams{
+		UniversityID: universityID,
+		Limit:        limit * 2, // Get more to filter
+	}
+
+	result, err := h.repo.SearchStaff(c.Request.Context(), params)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	// Filter to only include profiles that need extraction
+	var tasks []ProfileExtractionTask
+	for _, staff := range result.Staff {
+		// Skip if no profile URL
+		if staff.ProfileURL == nil || *staff.ProfileURL == "" {
+			continue
+		}
+
+		// Include if missing email or other important data
+		needsExtraction := staff.Email == nil || *staff.Email == ""
+
+		if needsExtraction {
+			task := ProfileExtractionTask{
+				StaffID:      staff.ID,
+				ProfileURL:   *staff.ProfileURL,
+				UniversityID: staff.UniversityID,
+			}
+
+			if staff.FullName != nil {
+				task.FullName = *staff.FullName
+			}
+			if staff.Email != nil {
+				task.CurrentData.Email = *staff.Email
+			}
+			if staff.Phone != nil {
+				task.CurrentData.Phone = *staff.Phone
+			}
+			if staff.Office != nil {
+				task.CurrentData.Office = *staff.Office
+			}
+			if staff.Position != nil {
+				task.CurrentData.Position = *staff.Position
+			}
+			if staff.DepartmentName != nil {
+				task.CurrentData.Department = *staff.DepartmentName
+			}
+
+			tasks = append(tasks, task)
+			if len(tasks) >= limit {
+				break
+			}
+		}
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"tasks": tasks,
+		"total": len(tasks),
+	})
+}
+
+// ExtractedProfileData represents data extracted by AI from a profile page
+type ExtractedProfileData struct {
+	StaffID uuid.UUID `json:"staff_id" binding:"required"`
+
+	// Contact info
+	Email string `json:"email,omitempty"`
+	Phone string `json:"phone,omitempty"`
+	Office string `json:"office,omitempty"`
+
+	// Professional info
+	Position       string `json:"position,omitempty"`
+	PositionType   string `json:"position_type,omitempty"` // professor, researcher, phd_student, staff
+	AcademicTitle  string `json:"academic_title,omitempty"`
+	IsProfessor    *bool  `json:"is_professor,omitempty"`
+	DepartmentName string `json:"department_name,omitempty"`
+
+	// Hierarchy
+	SupervisorName string `json:"supervisor_name,omitempty"`
+	TeamRole       string `json:"team_role,omitempty"` // leitung, mitarbeiter, sekretariat, hiwi, doktorand
+
+	// Research
+	ResearchInterests []string `json:"research_interests,omitempty"`
+	ResearchSummary   string   `json:"research_summary,omitempty"`
+
+	// Teaching (Lehrveranstaltungen)
+	TeachingTopics []string `json:"teaching_topics,omitempty"`
+
+	// External profiles
+	ORCID           string `json:"orcid,omitempty"`
+	GoogleScholarID string `json:"google_scholar_id,omitempty"`
+	ResearchgateURL string `json:"researchgate_url,omitempty"`
+	LinkedInURL     string `json:"linkedin_url,omitempty"`
+	PersonalWebsite string `json:"personal_website,omitempty"`
+	PhotoURL        string `json:"photo_url,omitempty"`
+
+	// Institute/Department links discovered
+	InstituteURL  string `json:"institute_url,omitempty"`
+	InstituteName string `json:"institute_name,omitempty"`
+
+	// Confidence score (0-1)
+	Confidence float64 `json:"confidence,omitempty"`
+}
+
+// SubmitExtractedData saves AI-extracted profile data
+// POST /api/v1/ai/extraction/submit
+func (h *AIExtractionHandlers) SubmitExtractedData(c *gin.Context) {
+	var data ExtractedProfileData
+	if err := c.ShouldBindJSON(&data); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request: " + err.Error()})
+		return
+	}
+
+	// Get existing staff record
+	staff, err := h.repo.GetStaff(c.Request.Context(), data.StaffID)
+	if err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Staff not found"})
+		return
+	}
+
+	// Update fields if provided and not empty
+	updated := false
+
+	if data.Email != "" && (staff.Email == nil || *staff.Email == "") {
+		staff.Email = &data.Email
+		updated = true
+	}
+	if data.Phone != "" && (staff.Phone == nil || *staff.Phone == "") {
+		staff.Phone = &data.Phone
+		updated = true
+	}
+	if data.Office != "" && (staff.Office == nil || *staff.Office == "") {
+		staff.Office = &data.Office
+		updated = true
+	}
+	if data.Position != "" && (staff.Position == nil || *staff.Position == "") {
+		staff.Position = &data.Position
+		updated = true
+	}
+	if data.PositionType != "" && (staff.PositionType == nil || *staff.PositionType == "") {
+		staff.PositionType = &data.PositionType
+		updated = true
+	}
+	if data.AcademicTitle != "" && (staff.AcademicTitle == nil || *staff.AcademicTitle == "") {
+		staff.AcademicTitle = &data.AcademicTitle
+		updated = true
+	}
+	if data.IsProfessor != nil {
+		staff.IsProfessor = *data.IsProfessor
+		updated = true
+	}
+	if data.TeamRole != "" && (staff.TeamRole == nil || *staff.TeamRole == "") {
+		staff.TeamRole = &data.TeamRole
+		updated = true
+	}
+	if len(data.ResearchInterests) > 0 && len(staff.ResearchInterests) == 0 {
+		staff.ResearchInterests = data.ResearchInterests
+		updated = true
+	}
+	if data.ResearchSummary != "" && (staff.ResearchSummary == nil || *staff.ResearchSummary == "") {
+		staff.ResearchSummary = &data.ResearchSummary
+		updated = true
+	}
+	if data.ORCID != "" && (staff.ORCID == nil || *staff.ORCID == "") {
+		staff.ORCID = &data.ORCID
+		updated = true
+	}
+	if data.GoogleScholarID != "" && (staff.GoogleScholarID == nil || *staff.GoogleScholarID == "") {
+		staff.GoogleScholarID = &data.GoogleScholarID
+		updated = true
+	}
+	if data.ResearchgateURL != "" && (staff.ResearchgateURL == nil || *staff.ResearchgateURL == "") {
+		staff.ResearchgateURL = &data.ResearchgateURL
+		updated = true
+	}
+	if data.LinkedInURL != "" && (staff.LinkedInURL == nil || *staff.LinkedInURL == "") {
+		staff.LinkedInURL = &data.LinkedInURL
+		updated = true
+	}
+	if data.PersonalWebsite != "" && (staff.PersonalWebsite == nil || *staff.PersonalWebsite == "") {
+		staff.PersonalWebsite = &data.PersonalWebsite
+		updated = true
+	}
+	if data.PhotoURL != "" && (staff.PhotoURL == nil || *staff.PhotoURL == "") {
+		staff.PhotoURL = &data.PhotoURL
+		updated = true
+	}
+
+	// Try to resolve supervisor by name
+	if data.SupervisorName != "" && staff.SupervisorID == nil {
+		// Search for supervisor in same university
+		supervisorParams := database.StaffSearchParams{
+			Query:        data.SupervisorName,
+			UniversityID: &staff.UniversityID,
+			Limit:        1,
+		}
+		result, err := h.repo.SearchStaff(c.Request.Context(), supervisorParams)
+		if err == nil && len(result.Staff) > 0 {
+			staff.SupervisorID = &result.Staff[0].ID
+			updated = true
+		}
+	}
+
+	// Update last verified timestamp
+	now := time.Now()
+	staff.LastVerified = &now
+
+	if updated {
+		err = h.repo.CreateStaff(c.Request.Context(), staff)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to update: " + err.Error()})
+			return
+		}
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"status":   "success",
+		"updated":  updated,
+		"staff_id": staff.ID,
+	})
+}
+
+// SubmitBatchExtractedData saves multiple AI-extracted profile data items
+// POST /api/v1/ai/extraction/submit-batch
+func (h *AIExtractionHandlers) SubmitBatchExtractedData(c *gin.Context) {
+	var batch struct {
+		Items []ExtractedProfileData `json:"items" binding:"required"`
+	}
+
+	if err := c.ShouldBindJSON(&batch); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request: " + err.Error()})
+		return
+	}
+
+	results := make([]gin.H, 0, len(batch.Items))
+	successCount := 0
+	errorCount := 0
+
+	for _, item := range batch.Items {
+		// Get existing staff record
+		staff, err := h.repo.GetStaff(c.Request.Context(), item.StaffID)
+		if err != nil {
+			results = append(results, gin.H{
+				"staff_id": item.StaffID,
+				"status":   "error",
+				"error":    "Staff not found",
+			})
+			errorCount++
+			continue
+		}
+
+		// Apply updates (same logic as single submit)
+		updated := false
+
+		if item.Email != "" && (staff.Email == nil || *staff.Email == "") {
+			staff.Email = &item.Email
+			updated = true
+		}
+		if item.Phone != "" && (staff.Phone == nil || *staff.Phone == "") {
+			staff.Phone = &item.Phone
+			updated = true
+		}
+		if item.Office != "" && (staff.Office == nil || *staff.Office == "") {
+			staff.Office = &item.Office
+			updated = true
+		}
+		if item.Position != "" && (staff.Position == nil || *staff.Position == "") {
+			staff.Position = &item.Position
+			updated = true
+		}
+		if item.PositionType != "" && (staff.PositionType == nil || *staff.PositionType == "") {
+			staff.PositionType = &item.PositionType
+			updated = true
+		}
+		if item.TeamRole != "" && (staff.TeamRole == nil || *staff.TeamRole == "") {
+			staff.TeamRole = &item.TeamRole
+			updated = true
+		}
+		if len(item.ResearchInterests) > 0 && len(staff.ResearchInterests) == 0 {
+			staff.ResearchInterests = item.ResearchInterests
+			updated = true
+		}
+		if item.ORCID != "" && (staff.ORCID == nil || *staff.ORCID == "") {
+			staff.ORCID = &item.ORCID
+			updated = true
+		}
+
+		// Update last verified
+		now := time.Now()
+		staff.LastVerified = &now
+
+		if updated {
+			err = h.repo.CreateStaff(c.Request.Context(), staff)
+			if err != nil {
+				results = append(results, gin.H{
+					"staff_id": item.StaffID,
+					"status":   "error",
+					"error":    err.Error(),
+				})
+				errorCount++
+				continue
+			}
+		}
+
+		results = append(results, gin.H{
+			"staff_id": item.StaffID,
+			"status":   "success",
+			"updated":  updated,
+		})
+		successCount++
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"results":       results,
+		"success_count": successCount,
+		"error_count":   errorCount,
+		"total":         len(batch.Items),
+	})
+}
+
+// InstituteHierarchyTask represents an institute page to crawl for hierarchy
+type InstituteHierarchyTask struct {
+	InstituteURL  string    `json:"institute_url"`
+	InstituteName string    `json:"institute_name,omitempty"`
+	UniversityID  uuid.UUID `json:"university_id"`
+}
+
+// GetInstitutePages returns institute pages that need hierarchy crawling
+// GET /api/v1/ai/extraction/institutes?university_id=...
+func (h *AIExtractionHandlers) GetInstitutePages(c *gin.Context) {
+	var universityID *uuid.UUID
+	if uniIDStr := c.Query("university_id"); uniIDStr != "" {
+		id, err := uuid.Parse(uniIDStr)
+		if err == nil {
+			universityID = &id
+		}
+	}
+
+	// Get unique institute/department URLs from staff profiles
+	params := database.StaffSearchParams{
+		UniversityID: universityID,
+		Limit:        1000,
+	}
+
+	result, err := h.repo.SearchStaff(c.Request.Context(), params)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	// Collect unique source URLs (these are typically department pages)
+	urlSet := make(map[string]bool)
+	var tasks []InstituteHierarchyTask
+
+	for _, staff := range result.Staff {
+		if staff.SourceURL != nil && *staff.SourceURL != "" {
+			url := *staff.SourceURL
+			if !urlSet[url] {
+				urlSet[url] = true
+				tasks = append(tasks, InstituteHierarchyTask{
+					InstituteURL: url,
+					UniversityID: staff.UniversityID,
+				})
+			}
+		}
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"institutes": tasks,
+		"total":      len(tasks),
+	})
+}
+
+// InstituteHierarchyData represents hierarchy data extracted from an institute page
+type InstituteHierarchyData struct {
+	InstituteURL  string    `json:"institute_url" binding:"required"`
+	UniversityID  uuid.UUID `json:"university_id" binding:"required"`
+	InstituteName string    `json:"institute_name,omitempty"`
+
+	// Leadership
+	LeaderName  string `json:"leader_name,omitempty"`
+	LeaderTitle string `json:"leader_title,omitempty"` // e.g., "Professor", "Lehrstuhlinhaber"
+
+	// Staff organization
+	StaffGroups []struct {
+		Role    string   `json:"role"`     // e.g., "Leitung", "Wissenschaftliche Mitarbeiter", "Sekretariat"
+		Members []string `json:"members"`  // Names of people in this group
+	} `json:"staff_groups,omitempty"`
+
+	// Teaching info (Lehrveranstaltungen)
+	TeachingCourses []struct {
+		Title   string `json:"title"`
+		Teacher string `json:"teacher,omitempty"`
+	} `json:"teaching_courses,omitempty"`
+}
+
+// SubmitInstituteHierarchy saves hierarchy data from an institute page
+// POST /api/v1/ai/extraction/institutes/submit
+func (h *AIExtractionHandlers) SubmitInstituteHierarchy(c *gin.Context) {
+	var data InstituteHierarchyData
+	if err := c.ShouldBindJSON(&data); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request: " + err.Error()})
+		return
+	}
+
+	// Find or create department
+	dept := &database.Department{
+		UniversityID: data.UniversityID,
+		Name:         data.InstituteName,
+	}
+	if data.InstituteURL != "" {
+		dept.URL = &data.InstituteURL
+	}
+
+	err := h.repo.CreateDepartment(c.Request.Context(), dept)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to create department: " + err.Error()})
+		return
+	}
+
+	// Find leader and set as supervisor for all staff in this institute
+	var leaderID *uuid.UUID
+	if data.LeaderName != "" {
+		// Search for leader
+		leaderParams := database.StaffSearchParams{
+			Query:        data.LeaderName,
+			UniversityID: &data.UniversityID,
+			Limit:        1,
+		}
+		result, err := h.repo.SearchStaff(c.Request.Context(), leaderParams)
+		if err == nil && len(result.Staff) > 0 {
+			leaderID = &result.Staff[0].ID
+
+			// Update leader with department and role
+			leader := &result.Staff[0]
+			leader.DepartmentID = &dept.ID
+			roleLeitung := "leitung"
+			leader.TeamRole = &roleLeitung
+			leader.IsProfessor = true
+			if data.LeaderTitle != "" {
+				leader.AcademicTitle = &data.LeaderTitle
+			}
+			h.repo.CreateStaff(c.Request.Context(), leader)
+		}
+	}
+
+	// Process staff groups
+	updatedCount := 0
+	for _, group := range data.StaffGroups {
+		for _, memberName := range group.Members {
+			// Find staff member
+			memberParams := database.StaffSearchParams{
+				Query:        memberName,
+				UniversityID: &data.UniversityID,
+				Limit:        1,
+			}
+			result, err := h.repo.SearchStaff(c.Request.Context(), memberParams)
+			if err != nil || len(result.Staff) == 0 {
+				continue
+			}
+
+			member := &result.Staff[0]
+			member.DepartmentID = &dept.ID
+			member.TeamRole = &group.Role
+
+			// Set supervisor if leader was found and this is not the leader
+			if leaderID != nil && member.ID != *leaderID {
+				member.SupervisorID = leaderID
+			}
+
+			h.repo.CreateStaff(c.Request.Context(), member)
+			updatedCount++
+		}
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"status":          "success",
+		"department_id":   dept.ID,
+		"leader_id":       leaderID,
+		"members_updated": updatedCount,
+	})
+}
+
+// RegisterAIExtractionRoutes registers AI extraction routes
+func (h *AIExtractionHandlers) RegisterRoutes(r *gin.RouterGroup) {
+	ai := r.Group("/ai/extraction")
+
+	// Profile extraction endpoints
+	ai.GET("/pending", h.GetPendingProfiles)
+	ai.POST("/submit", h.SubmitExtractedData)
+	ai.POST("/submit-batch", h.SubmitBatchExtractedData)
+
+	// Institute hierarchy endpoints
+	ai.GET("/institutes", h.GetInstitutePages)
+	ai.POST("/institutes/submit", h.SubmitInstituteHierarchy)
+}
--- a/edu-search-service/internal/api/handlers/audience_handlers.go
+++ b/edu-search-service/internal/api/handlers/audience_handlers.go
@@ -0,0 +1,314 @@
+package handlers
+
+import (
+	"net/http"
+	"strconv"
+
+	"github.com/breakpilot/edu-search-service/internal/orchestrator"
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+)
+
+// AudienceHandler handles audience-related HTTP requests
+type AudienceHandler struct {
+	repo orchestrator.AudienceRepository
+}
+
+// NewAudienceHandler creates a new audience handler
+func NewAudienceHandler(repo orchestrator.AudienceRepository) *AudienceHandler {
+	return &AudienceHandler{repo: repo}
+}
+
+// CreateAudienceRequest represents a request to create an audience
+type CreateAudienceRequest struct {
+	Name        string                      `json:"name" binding:"required"`
+	Description string                      `json:"description"`
+	Filters     orchestrator.AudienceFilters `json:"filters"`
+	CreatedBy   string                      `json:"created_by"`
+}
+
+// UpdateAudienceRequest represents a request to update an audience
+type UpdateAudienceRequest struct {
+	Name        string                      `json:"name" binding:"required"`
+	Description string                      `json:"description"`
+	Filters     orchestrator.AudienceFilters `json:"filters"`
+	IsActive    bool                        `json:"is_active"`
+}
+
+// CreateExportRequest represents a request to create an export
+type CreateExportRequest struct {
+	ExportType string `json:"export_type" binding:"required"` // csv, json, email_list
+	Purpose    string `json:"purpose"`
+	ExportedBy string `json:"exported_by"`
+}
+
+// ListAudiences returns all audiences
+func (h *AudienceHandler) ListAudiences(c *gin.Context) {
+	activeOnly := c.Query("active_only") == "true"
+
+	audiences, err := h.repo.ListAudiences(c.Request.Context(), activeOnly)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to list audiences", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"audiences": audiences,
+		"count":     len(audiences),
+	})
+}
+
+// GetAudience returns a single audience
+func (h *AudienceHandler) GetAudience(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid audience ID"})
+		return
+	}
+
+	audience, err := h.repo.GetAudience(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Audience not found", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, audience)
+}
+
+// CreateAudience creates a new audience
+func (h *AudienceHandler) CreateAudience(c *gin.Context) {
+	var req CreateAudienceRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	audience := &orchestrator.Audience{
+		Name:        req.Name,
+		Description: req.Description,
+		Filters:     req.Filters,
+		CreatedBy:   req.CreatedBy,
+		IsActive:    true,
+	}
+
+	if err := h.repo.CreateAudience(c.Request.Context(), audience); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to create audience", "details": err.Error()})
+		return
+	}
+
+	// Update the member count
+	count, _ := h.repo.UpdateAudienceCount(c.Request.Context(), audience.ID)
+	audience.MemberCount = count
+
+	c.JSON(http.StatusCreated, audience)
+}
+
+// UpdateAudience updates an existing audience
+func (h *AudienceHandler) UpdateAudience(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid audience ID"})
+		return
+	}
+
+	var req UpdateAudienceRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	audience := &orchestrator.Audience{
+		ID:          id,
+		Name:        req.Name,
+		Description: req.Description,
+		Filters:     req.Filters,
+		IsActive:    req.IsActive,
+	}
+
+	if err := h.repo.UpdateAudience(c.Request.Context(), audience); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to update audience", "details": err.Error()})
+		return
+	}
+
+	// Update the member count
+	count, _ := h.repo.UpdateAudienceCount(c.Request.Context(), audience.ID)
+	audience.MemberCount = count
+
+	c.JSON(http.StatusOK, audience)
+}
+
+// DeleteAudience soft-deletes an audience
+func (h *AudienceHandler) DeleteAudience(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid audience ID"})
+		return
+	}
+
+	if err := h.repo.DeleteAudience(c.Request.Context(), id); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to delete audience", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{"deleted": true, "id": idStr})
+}
+
+// GetAudienceMembers returns members matching the audience filters
+func (h *AudienceHandler) GetAudienceMembers(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid audience ID"})
+		return
+	}
+
+	// Parse pagination
+	limit := 50
+	offset := 0
+	if l := c.Query("limit"); l != "" {
+		if parsed, err := strconv.Atoi(l); err == nil && parsed > 0 && parsed <= 500 {
+			limit = parsed
+		}
+	}
+	if o := c.Query("offset"); o != "" {
+		if parsed, err := strconv.Atoi(o); err == nil && parsed >= 0 {
+			offset = parsed
+		}
+	}
+
+	members, totalCount, err := h.repo.GetAudienceMembers(c.Request.Context(), id, limit, offset)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get members", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"members":     members,
+		"count":       len(members),
+		"total_count": totalCount,
+		"limit":       limit,
+		"offset":      offset,
+	})
+}
+
+// RefreshAudienceCount recalculates the member count
+func (h *AudienceHandler) RefreshAudienceCount(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid audience ID"})
+		return
+	}
+
+	count, err := h.repo.UpdateAudienceCount(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to refresh count", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"audience_id":  idStr,
+		"member_count": count,
+	})
+}
+
+// PreviewAudienceFilters previews the result of filters without saving
+func (h *AudienceHandler) PreviewAudienceFilters(c *gin.Context) {
+	var filters orchestrator.AudienceFilters
+	if err := c.ShouldBindJSON(&filters); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	// Return the filters for now - preview functionality can be expanded later
+	c.JSON(http.StatusOK, gin.H{
+		"filters": filters,
+		"message": "Preview functionality requires direct repository access",
+	})
+}
+
+// CreateExport creates a new export for an audience
+func (h *AudienceHandler) CreateExport(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid audience ID"})
+		return
+	}
+
+	var req CreateExportRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	// Get the member count for the export
+	_, totalCount, err := h.repo.GetAudienceMembers(c.Request.Context(), id, 1, 0)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get members", "details": err.Error()})
+		return
+	}
+
+	export := &orchestrator.AudienceExport{
+		AudienceID:  id,
+		ExportType:  req.ExportType,
+		RecordCount: totalCount,
+		ExportedBy:  req.ExportedBy,
+		Purpose:     req.Purpose,
+	}
+
+	if err := h.repo.CreateExport(c.Request.Context(), export); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to create export", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusCreated, export)
+}
+
+// ListExports lists exports for an audience
+func (h *AudienceHandler) ListExports(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid audience ID"})
+		return
+	}
+
+	exports, err := h.repo.ListExports(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to list exports", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"exports": exports,
+		"count":   len(exports),
+	})
+}
+
+// SetupAudienceRoutes configures audience API routes
+func SetupAudienceRoutes(r *gin.RouterGroup, h *AudienceHandler) {
+	audiences := r.Group("/audiences")
+	{
+		// Audience CRUD
+		audiences.GET("", h.ListAudiences)
+		audiences.GET("/:id", h.GetAudience)
+		audiences.POST("", h.CreateAudience)
+		audiences.PUT("/:id", h.UpdateAudience)
+		audiences.DELETE("/:id", h.DeleteAudience)
+
+		// Members
+		audiences.GET("/:id/members", h.GetAudienceMembers)
+		audiences.POST("/:id/refresh", h.RefreshAudienceCount)
+
+		// Exports
+		audiences.GET("/:id/exports", h.ListExports)
+		audiences.POST("/:id/exports", h.CreateExport)
+
+		// Preview (no audience required)
+		audiences.POST("/preview", h.PreviewAudienceFilters)
+	}
+}
--- a/edu-search-service/internal/api/handlers/audience_handlers_test.go
+++ b/edu-search-service/internal/api/handlers/audience_handlers_test.go
@@ -0,0 +1,630 @@
+package handlers
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/breakpilot/edu-search-service/internal/orchestrator"
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+)
+
+// MockAudienceRepository implements orchestrator.AudienceRepository for testing
+type MockAudienceRepository struct {
+	audiences []orchestrator.Audience
+	exports   []orchestrator.AudienceExport
+	members   []orchestrator.AudienceMember
+}
+
+func NewMockAudienceRepository() *MockAudienceRepository {
+	return &MockAudienceRepository{
+		audiences: make([]orchestrator.Audience, 0),
+		exports:   make([]orchestrator.AudienceExport, 0),
+		members:   make([]orchestrator.AudienceMember, 0),
+	}
+}
+
+func (m *MockAudienceRepository) CreateAudience(ctx context.Context, audience *orchestrator.Audience) error {
+	audience.ID = uuid.New()
+	audience.CreatedAt = time.Now()
+	audience.UpdatedAt = time.Now()
+	m.audiences = append(m.audiences, *audience)
+	return nil
+}
+
+func (m *MockAudienceRepository) GetAudience(ctx context.Context, id uuid.UUID) (*orchestrator.Audience, error) {
+	for i := range m.audiences {
+		if m.audiences[i].ID == id {
+			return &m.audiences[i], nil
+		}
+	}
+	return nil, context.DeadlineExceeded // simulate not found
+}
+
+func (m *MockAudienceRepository) ListAudiences(ctx context.Context, activeOnly bool) ([]orchestrator.Audience, error) {
+	if activeOnly {
+		var active []orchestrator.Audience
+		for _, a := range m.audiences {
+			if a.IsActive {
+				active = append(active, a)
+			}
+		}
+		return active, nil
+	}
+	return m.audiences, nil
+}
+
+func (m *MockAudienceRepository) UpdateAudience(ctx context.Context, audience *orchestrator.Audience) error {
+	for i := range m.audiences {
+		if m.audiences[i].ID == audience.ID {
+			m.audiences[i].Name = audience.Name
+			m.audiences[i].Description = audience.Description
+			m.audiences[i].Filters = audience.Filters
+			m.audiences[i].IsActive = audience.IsActive
+			m.audiences[i].UpdatedAt = time.Now()
+			audience.UpdatedAt = m.audiences[i].UpdatedAt
+			return nil
+		}
+	}
+	return nil
+}
+
+func (m *MockAudienceRepository) DeleteAudience(ctx context.Context, id uuid.UUID) error {
+	for i := range m.audiences {
+		if m.audiences[i].ID == id {
+			m.audiences[i].IsActive = false
+			return nil
+		}
+	}
+	return nil
+}
+
+func (m *MockAudienceRepository) GetAudienceMembers(ctx context.Context, id uuid.UUID, limit, offset int) ([]orchestrator.AudienceMember, int, error) {
+	// Return mock members
+	if len(m.members) == 0 {
+		m.members = []orchestrator.AudienceMember{
+			{
+				ID:               uuid.New(),
+				Name:             "Prof. Dr. Test Person",
+				Email:            "test@university.de",
+				Position:         "professor",
+				University:       "Test Universität",
+				Department:       "Informatik",
+				SubjectArea:      "Informatik",
+				PublicationCount: 42,
+			},
+			{
+				ID:               uuid.New(),
+				Name:             "Dr. Another Person",
+				Email:            "another@university.de",
+				Position:         "researcher",
+				University:       "Test Universität",
+				Department:       "Mathematik",
+				SubjectArea:      "Mathematik",
+				PublicationCount: 15,
+			},
+		}
+	}
+
+	total := len(m.members)
+	if offset >= total {
+		return []orchestrator.AudienceMember{}, total, nil
+	}
+
+	end := offset + limit
+	if end > total {
+		end = total
+	}
+
+	return m.members[offset:end], total, nil
+}
+
+func (m *MockAudienceRepository) UpdateAudienceCount(ctx context.Context, id uuid.UUID) (int, error) {
+	count := len(m.members)
+	for i := range m.audiences {
+		if m.audiences[i].ID == id {
+			m.audiences[i].MemberCount = count
+			now := time.Now()
+			m.audiences[i].LastCountUpdate = &now
+		}
+	}
+	return count, nil
+}
+
+func (m *MockAudienceRepository) CreateExport(ctx context.Context, export *orchestrator.AudienceExport) error {
+	export.ID = uuid.New()
+	export.CreatedAt = time.Now()
+	m.exports = append(m.exports, *export)
+	return nil
+}
+
+func (m *MockAudienceRepository) ListExports(ctx context.Context, audienceID uuid.UUID) ([]orchestrator.AudienceExport, error) {
+	var exports []orchestrator.AudienceExport
+	for _, e := range m.exports {
+		if e.AudienceID == audienceID {
+			exports = append(exports, e)
+		}
+	}
+	return exports, nil
+}
+
+func setupAudienceRouter(repo *MockAudienceRepository) *gin.Engine {
+	gin.SetMode(gin.TestMode)
+	router := gin.New()
+
+	handler := NewAudienceHandler(repo)
+
+	v1 := router.Group("/v1")
+	SetupAudienceRoutes(v1, handler)
+
+	return router
+}
+
+func TestAudienceHandler_ListAudiences_Empty(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var response struct {
+		Audiences []orchestrator.Audience `json:"audiences"`
+		Count     int                     `json:"count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.Count != 0 {
+		t.Errorf("Expected 0 audiences, got %d", response.Count)
+	}
+}
+
+func TestAudienceHandler_CreateAudience(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	body := CreateAudienceRequest{
+		Name:        "Test Audience",
+		Description: "A test audience for professors",
+		Filters: orchestrator.AudienceFilters{
+			PositionTypes: []string{"professor"},
+			States:        []string{"BW", "BY"},
+		},
+		CreatedBy: "test-admin",
+	}
+
+	bodyJSON, _ := json.Marshal(body)
+	req := httptest.NewRequest(http.MethodPost, "/v1/audiences", bytes.NewBuffer(bodyJSON))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusCreated {
+		t.Errorf("Expected status %d, got %d: %s", http.StatusCreated, w.Code, w.Body.String())
+	}
+
+	var response orchestrator.Audience
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.Name != "Test Audience" {
+		t.Errorf("Expected name 'Test Audience', got '%s'", response.Name)
+	}
+
+	if !response.IsActive {
+		t.Errorf("Expected audience to be active")
+	}
+
+	if len(repo.audiences) != 1 {
+		t.Errorf("Expected 1 audience in repo, got %d", len(repo.audiences))
+	}
+}
+
+func TestAudienceHandler_CreateAudience_InvalidJSON(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/audiences", bytes.NewBuffer([]byte("invalid json")))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status %d, got %d", http.StatusBadRequest, w.Code)
+	}
+}
+
+func TestAudienceHandler_CreateAudience_MissingName(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	body := map[string]interface{}{
+		"description": "Missing name field",
+	}
+
+	bodyJSON, _ := json.Marshal(body)
+	req := httptest.NewRequest(http.MethodPost, "/v1/audiences", bytes.NewBuffer(bodyJSON))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status %d, got %d", http.StatusBadRequest, w.Code)
+	}
+}
+
+func TestAudienceHandler_GetAudience(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	// Create an audience first
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "Test Audience",
+		Description: "Test description",
+		IsActive:    true,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences/"+audience.ID.String(), nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d: %s", http.StatusOK, w.Code, w.Body.String())
+	}
+
+	var response orchestrator.Audience
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.Name != "Test Audience" {
+		t.Errorf("Expected name 'Test Audience', got '%s'", response.Name)
+	}
+}
+
+func TestAudienceHandler_GetAudience_InvalidID(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences/invalid-uuid", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status %d, got %d", http.StatusBadRequest, w.Code)
+	}
+}
+
+func TestAudienceHandler_GetAudience_NotFound(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences/"+uuid.New().String(), nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusNotFound {
+		t.Errorf("Expected status %d, got %d", http.StatusNotFound, w.Code)
+	}
+}
+
+func TestAudienceHandler_UpdateAudience(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	// Create an audience first
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "Old Name",
+		Description: "Old description",
+		IsActive:    true,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	body := UpdateAudienceRequest{
+		Name:        "New Name",
+		Description: "New description",
+		IsActive:    true,
+	}
+
+	bodyJSON, _ := json.Marshal(body)
+	req := httptest.NewRequest(http.MethodPut, "/v1/audiences/"+audience.ID.String(), bytes.NewBuffer(bodyJSON))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d: %s", http.StatusOK, w.Code, w.Body.String())
+	}
+
+	// Verify the update
+	if repo.audiences[0].Name != "New Name" {
+		t.Errorf("Expected name 'New Name', got '%s'", repo.audiences[0].Name)
+	}
+}
+
+func TestAudienceHandler_DeleteAudience(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	// Create an audience first
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "To Delete",
+		IsActive:    true,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	req := httptest.NewRequest(http.MethodDelete, "/v1/audiences/"+audience.ID.String(), nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	// Verify soft delete
+	if repo.audiences[0].IsActive {
+		t.Errorf("Expected audience to be inactive after delete")
+	}
+}
+
+func TestAudienceHandler_GetAudienceMembers(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	// Create an audience first
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "Test Audience",
+		IsActive:    true,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences/"+audience.ID.String()+"/members", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d: %s", http.StatusOK, w.Code, w.Body.String())
+	}
+
+	var response struct {
+		Members    []orchestrator.AudienceMember `json:"members"`
+		Count      int                           `json:"count"`
+		TotalCount int                           `json:"total_count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.TotalCount != 2 {
+		t.Errorf("Expected 2 total members, got %d", response.TotalCount)
+	}
+}
+
+func TestAudienceHandler_GetAudienceMembers_WithPagination(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "Test Audience",
+		IsActive:    true,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences/"+audience.ID.String()+"/members?limit=1&offset=0", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var response struct {
+		Members []orchestrator.AudienceMember `json:"members"`
+		Count   int                           `json:"count"`
+		Limit   int                           `json:"limit"`
+		Offset  int                           `json:"offset"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.Count != 1 {
+		t.Errorf("Expected 1 member in response, got %d", response.Count)
+	}
+
+	if response.Limit != 1 {
+		t.Errorf("Expected limit 1, got %d", response.Limit)
+	}
+}
+
+func TestAudienceHandler_RefreshAudienceCount(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "Test Audience",
+		IsActive:    true,
+		MemberCount: 0,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	// Pre-initialize members so count works correctly
+	repo.members = []orchestrator.AudienceMember{
+		{ID: uuid.New(), Name: "Test Person 1"},
+		{ID: uuid.New(), Name: "Test Person 2"},
+	}
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/audiences/"+audience.ID.String()+"/refresh", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var response struct {
+		AudienceID  string `json:"audience_id"`
+		MemberCount int    `json:"member_count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.MemberCount != 2 {
+		t.Errorf("Expected member_count 2, got %d", response.MemberCount)
+	}
+}
+
+func TestAudienceHandler_CreateExport(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "Test Audience",
+		IsActive:    true,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	body := CreateExportRequest{
+		ExportType: "csv",
+		Purpose:    "Newsletter December 2024",
+		ExportedBy: "admin",
+	}
+
+	bodyJSON, _ := json.Marshal(body)
+	req := httptest.NewRequest(http.MethodPost, "/v1/audiences/"+audience.ID.String()+"/exports", bytes.NewBuffer(bodyJSON))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusCreated {
+		t.Errorf("Expected status %d, got %d: %s", http.StatusCreated, w.Code, w.Body.String())
+	}
+
+	var response orchestrator.AudienceExport
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.ExportType != "csv" {
+		t.Errorf("Expected export_type 'csv', got '%s'", response.ExportType)
+	}
+
+	if response.RecordCount != 2 {
+		t.Errorf("Expected record_count 2, got %d", response.RecordCount)
+	}
+}
+
+func TestAudienceHandler_ListExports(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	audience := orchestrator.Audience{
+		ID:          uuid.New(),
+		Name:        "Test Audience",
+		IsActive:    true,
+		CreatedAt:   time.Now(),
+		UpdatedAt:   time.Now(),
+	}
+	repo.audiences = append(repo.audiences, audience)
+
+	// Add an export
+	export := orchestrator.AudienceExport{
+		ID:          uuid.New(),
+		AudienceID:  audience.ID,
+		ExportType:  "csv",
+		RecordCount: 100,
+		Purpose:     "Test export",
+		CreatedAt:   time.Now(),
+	}
+	repo.exports = append(repo.exports, export)
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences/"+audience.ID.String()+"/exports", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var response struct {
+		Exports []orchestrator.AudienceExport `json:"exports"`
+		Count   int                           `json:"count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.Count != 1 {
+		t.Errorf("Expected 1 export, got %d", response.Count)
+	}
+}
+
+func TestAudienceHandler_ListAudiences_ActiveOnly(t *testing.T) {
+	repo := NewMockAudienceRepository()
+	router := setupAudienceRouter(repo)
+
+	// Add active and inactive audiences
+	repo.audiences = []orchestrator.Audience{
+		{ID: uuid.New(), Name: "Active", IsActive: true, CreatedAt: time.Now(), UpdatedAt: time.Now()},
+		{ID: uuid.New(), Name: "Inactive", IsActive: false, CreatedAt: time.Now(), UpdatedAt: time.Now()},
+	}
+
+	req := httptest.NewRequest(http.MethodGet, "/v1/audiences?active_only=true", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status %d, got %d", http.StatusOK, w.Code)
+	}
+
+	var response struct {
+		Audiences []orchestrator.Audience `json:"audiences"`
+		Count     int                     `json:"count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to unmarshal response: %v", err)
+	}
+
+	if response.Count != 1 {
+		t.Errorf("Expected 1 active audience, got %d", response.Count)
+	}
+
+	if response.Audiences[0].Name != "Active" {
+		t.Errorf("Expected audience 'Active', got '%s'", response.Audiences[0].Name)
+	}
+}
--- a/edu-search-service/internal/api/handlers/handlers.go
+++ b/edu-search-service/internal/api/handlers/handlers.go
@@ -0,0 +1,146 @@
+package handlers
+
+import (
+	"net/http"
+
+	"github.com/breakpilot/edu-search-service/internal/config"
+	"github.com/breakpilot/edu-search-service/internal/indexer"
+	"github.com/breakpilot/edu-search-service/internal/search"
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+)
+
+// Handler contains all HTTP handlers
+type Handler struct {
+	cfg           *config.Config
+	searchService *search.Service
+	indexClient   *indexer.Client
+}
+
+// NewHandler creates a new handler instance
+func NewHandler(cfg *config.Config, searchService *search.Service, indexClient *indexer.Client) *Handler {
+	return &Handler{
+		cfg:           cfg,
+		searchService: searchService,
+		indexClient:   indexClient,
+	}
+}
+
+// Health returns service health status
+func (h *Handler) Health(c *gin.Context) {
+	status := "ok"
+
+	// Check OpenSearch health
+	osStatus, err := h.indexClient.Health(c.Request.Context())
+	if err != nil {
+		status = "degraded"
+		osStatus = "unreachable"
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"status":           status,
+		"opensearch":       osStatus,
+		"service":          "edu-search-service",
+		"version":          "0.1.0",
+	})
+}
+
+// Search handles /v1/search requests
+func (h *Handler) Search(c *gin.Context) {
+	var req search.SearchRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	// Set defaults
+	if req.Limit <= 0 || req.Limit > 100 {
+		req.Limit = 10
+	}
+	if req.Mode == "" {
+		req.Mode = "keyword" // MVP: only BM25
+	}
+
+	// Generate query ID
+	queryID := uuid.New().String()
+
+	// Execute search
+	result, err := h.searchService.Search(c.Request.Context(), &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Search failed", "details": err.Error()})
+		return
+	}
+
+	result.QueryID = queryID
+	c.JSON(http.StatusOK, result)
+}
+
+// GetDocument retrieves a single document
+func (h *Handler) GetDocument(c *gin.Context) {
+	docID := c.Query("doc_id")
+	if docID == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "doc_id parameter required"})
+		return
+	}
+
+	// TODO: Implement document retrieval
+	c.JSON(http.StatusNotImplemented, gin.H{"error": "Not implemented yet"})
+}
+
+// AuthMiddleware validates API keys
+func AuthMiddleware(apiKey string) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		// Skip auth for health endpoint
+		if c.Request.URL.Path == "/v1/health" {
+			c.Next()
+			return
+		}
+
+		// Check API key
+		authHeader := c.GetHeader("Authorization")
+		if authHeader == "" {
+			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{"error": "Missing Authorization header"})
+			return
+		}
+
+		// Extract Bearer token
+		if len(authHeader) < 7 || authHeader[:7] != "Bearer " {
+			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{"error": "Invalid Authorization format"})
+			return
+		}
+
+		token := authHeader[7:]
+		if apiKey != "" && token != apiKey {
+			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{"error": "Invalid API key"})
+			return
+		}
+
+		c.Next()
+	}
+}
+
+// RateLimitMiddleware implements basic rate limiting
+func RateLimitMiddleware() gin.HandlerFunc {
+	// TODO: Implement proper rate limiting with Redis
+	return func(c *gin.Context) {
+		c.Next()
+	}
+}
+
+// SetupRoutes configures all API routes
+func SetupRoutes(r *gin.Engine, h *Handler, apiKey string) {
+	// Health endpoint (no auth)
+	r.GET("/v1/health", h.Health)
+
+	// API v1 group with auth
+	v1 := r.Group("/v1")
+	v1.Use(AuthMiddleware(apiKey))
+	v1.Use(RateLimitMiddleware())
+	{
+		v1.POST("/search", h.Search)
+		v1.GET("/document", h.GetDocument)
+
+		// Admin routes
+		SetupAdminRoutes(v1, h)
+	}
+}
--- a/edu-search-service/internal/api/handlers/handlers_test.go
+++ b/edu-search-service/internal/api/handlers/handlers_test.go
@@ -0,0 +1,645 @@
+package handlers
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+)
+
+func init() {
+	gin.SetMode(gin.TestMode)
+}
+
+// setupTestRouter creates a test router with the handler
+func setupTestRouter(h *Handler, apiKey string) *gin.Engine {
+	router := gin.New()
+	SetupRoutes(router, h, apiKey)
+	return router
+}
+
+// setupTestSeedStore creates a test seed store
+func setupTestSeedStore(t *testing.T) string {
+	t.Helper()
+	dir := t.TempDir()
+
+	// Initialize global seed store
+	err := InitSeedStore(dir)
+	if err != nil {
+		t.Fatalf("Failed to initialize seed store: %v", err)
+	}
+
+	return dir
+}
+
+func TestHealthEndpoint(t *testing.T) {
+	// Health endpoint requires indexClient for health check
+	// This test verifies the route is set up correctly
+	// A full integration test would need a mock OpenSearch client
+	t.Skip("Skipping: requires mock indexer client for full test")
+}
+
+func TestAuthMiddleware_NoAuth(t *testing.T) {
+	h := &Handler{}
+	router := setupTestRouter(h, "test-api-key")
+
+	// Request without auth header
+	req, _ := http.NewRequest("POST", "/v1/search", bytes.NewBufferString(`{"q":"test"}`))
+	req.Header.Set("Content-Type", "application/json")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("Expected status 401, got %d", w.Code)
+	}
+}
+
+func TestAuthMiddleware_InvalidFormat(t *testing.T) {
+	h := &Handler{}
+	router := setupTestRouter(h, "test-api-key")
+
+	// Request with wrong auth format
+	req, _ := http.NewRequest("POST", "/v1/search", bytes.NewBufferString(`{"q":"test"}`))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Basic dGVzdDp0ZXN0") // Basic auth instead of Bearer
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("Expected status 401, got %d", w.Code)
+	}
+}
+
+func TestAuthMiddleware_InvalidKey(t *testing.T) {
+	h := &Handler{}
+	router := setupTestRouter(h, "test-api-key")
+
+	// Request with wrong API key
+	req, _ := http.NewRequest("POST", "/v1/search", bytes.NewBufferString(`{"q":"test"}`))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer wrong-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("Expected status 401, got %d", w.Code)
+	}
+}
+
+func TestAuthMiddleware_ValidKey(t *testing.T) {
+	h := &Handler{}
+	router := setupTestRouter(h, "test-api-key")
+
+	// Request with correct API key (search will fail due to no search service, but auth should pass)
+	req, _ := http.NewRequest("GET", "/v1/document?doc_id=test", nil)
+	req.Header.Set("Authorization", "Bearer test-api-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	// Auth should pass, endpoint returns 501 (not implemented)
+	if w.Code == http.StatusUnauthorized {
+		t.Error("Expected auth to pass, got 401")
+	}
+}
+
+func TestAuthMiddleware_HealthNoAuth(t *testing.T) {
+	// Health endpoint requires indexClient for health check
+	// Skipping because route calls h.indexClient.Health() which panics with nil
+	t.Skip("Skipping: requires mock indexer client for full test")
+}
+
+func TestGetDocument_MissingDocID(t *testing.T) {
+	h := &Handler{}
+	router := setupTestRouter(h, "test-key")
+
+	req, _ := http.NewRequest("GET", "/v1/document", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400, got %d", w.Code)
+	}
+}
+
+// Admin Handler Tests
+
+func TestSeedStore_InitAndLoad(t *testing.T) {
+	dir := t.TempDir()
+
+	// First initialization should create default seeds
+	err := InitSeedStore(dir)
+	if err != nil {
+		t.Fatalf("InitSeedStore failed: %v", err)
+	}
+
+	// Check that seeds file was created
+	seedsFile := filepath.Join(dir, "seeds.json")
+	if _, err := os.Stat(seedsFile); os.IsNotExist(err) {
+		t.Error("seeds.json was not created")
+	}
+
+	// Check that default seeds were loaded
+	seeds := seedStore.GetAllSeeds()
+	if len(seeds) == 0 {
+		t.Error("Expected default seeds to be loaded")
+	}
+}
+
+func TestSeedStore_CreateSeed(t *testing.T) {
+	setupTestSeedStore(t)
+
+	newSeed := SeedURL{
+		URL:         "https://test.example.com",
+		Name:        "Test Seed",
+		Category:    "test",
+		Description: "A test seed",
+		TrustBoost:  0.5,
+		Enabled:     true,
+	}
+
+	created, err := seedStore.CreateSeed(newSeed)
+	if err != nil {
+		t.Fatalf("CreateSeed failed: %v", err)
+	}
+
+	if created.ID == "" {
+		t.Error("Expected generated ID")
+	}
+	if created.URL != newSeed.URL {
+		t.Errorf("Expected URL %q, got %q", newSeed.URL, created.URL)
+	}
+	if created.CreatedAt.IsZero() {
+		t.Error("Expected CreatedAt to be set")
+	}
+}
+
+func TestSeedStore_GetSeed(t *testing.T) {
+	setupTestSeedStore(t)
+
+	// Create a seed first
+	newSeed := SeedURL{
+		URL:      "https://get-test.example.com",
+		Name:     "Get Test",
+		Category: "test",
+	}
+	created, _ := seedStore.CreateSeed(newSeed)
+
+	// Get the seed
+	retrieved, found := seedStore.GetSeed(created.ID)
+	if !found {
+		t.Fatal("Seed not found")
+	}
+
+	if retrieved.URL != newSeed.URL {
+		t.Errorf("Expected URL %q, got %q", newSeed.URL, retrieved.URL)
+	}
+}
+
+func TestSeedStore_GetSeed_NotFound(t *testing.T) {
+	setupTestSeedStore(t)
+
+	_, found := seedStore.GetSeed("nonexistent-id")
+	if found {
+		t.Error("Expected seed not to be found")
+	}
+}
+
+func TestSeedStore_UpdateSeed(t *testing.T) {
+	setupTestSeedStore(t)
+
+	// Create a seed first
+	original := SeedURL{
+		URL:      "https://update-test.example.com",
+		Name:     "Original Name",
+		Category: "test",
+		Enabled:  true,
+	}
+	created, _ := seedStore.CreateSeed(original)
+
+	// Update the seed
+	updates := SeedURL{
+		Name:       "Updated Name",
+		TrustBoost: 0.75,
+		Enabled:    false,
+	}
+
+	updated, found, err := seedStore.UpdateSeed(created.ID, updates)
+	if err != nil {
+		t.Fatalf("UpdateSeed failed: %v", err)
+	}
+	if !found {
+		t.Fatal("Seed not found for update")
+	}
+
+	if updated.Name != "Updated Name" {
+		t.Errorf("Expected name 'Updated Name', got %q", updated.Name)
+	}
+	if updated.TrustBoost != 0.75 {
+		t.Errorf("Expected TrustBoost 0.75, got %f", updated.TrustBoost)
+	}
+	if updated.Enabled != false {
+		t.Error("Expected Enabled to be false")
+	}
+	// URL should remain unchanged since we didn't provide it
+	if updated.URL != original.URL {
+		t.Errorf("URL should remain unchanged, expected %q, got %q", original.URL, updated.URL)
+	}
+}
+
+func TestSeedStore_UpdateSeed_NotFound(t *testing.T) {
+	setupTestSeedStore(t)
+
+	updates := SeedURL{Name: "New Name"}
+	_, found, err := seedStore.UpdateSeed("nonexistent-id", updates)
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+	if found {
+		t.Error("Expected seed not to be found")
+	}
+}
+
+func TestSeedStore_DeleteSeed(t *testing.T) {
+	setupTestSeedStore(t)
+
+	// Create a seed first
+	newSeed := SeedURL{
+		URL:      "https://delete-test.example.com",
+		Name:     "Delete Test",
+		Category: "test",
+	}
+	created, _ := seedStore.CreateSeed(newSeed)
+
+	// Delete the seed
+	deleted := seedStore.DeleteSeed(created.ID)
+	if !deleted {
+		t.Error("Expected delete to succeed")
+	}
+
+	// Verify it's gone
+	_, found := seedStore.GetSeed(created.ID)
+	if found {
+		t.Error("Seed should have been deleted")
+	}
+}
+
+func TestSeedStore_DeleteSeed_NotFound(t *testing.T) {
+	setupTestSeedStore(t)
+
+	deleted := seedStore.DeleteSeed("nonexistent-id")
+	if deleted {
+		t.Error("Expected delete to return false for nonexistent seed")
+	}
+}
+
+func TestSeedStore_Persistence(t *testing.T) {
+	dir := t.TempDir()
+
+	// Create and populate seed store
+	err := InitSeedStore(dir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	newSeed := SeedURL{
+		URL:      "https://persist-test.example.com",
+		Name:     "Persistence Test",
+		Category: "test",
+	}
+	created, err := seedStore.CreateSeed(newSeed)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Re-initialize from the same directory
+	seedStore = nil
+	err = InitSeedStore(dir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Check if the seed persisted
+	retrieved, found := seedStore.GetSeed(created.ID)
+	if !found {
+		t.Error("Seed should have persisted")
+	}
+	if retrieved.URL != newSeed.URL {
+		t.Errorf("Persisted seed URL mismatch: expected %q, got %q", newSeed.URL, retrieved.URL)
+	}
+}
+
+func TestAdminGetSeeds(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+
+	// Initialize seed store for the test
+	InitSeedStore(dir)
+
+	req, _ := http.NewRequest("GET", "/v1/admin/seeds", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d", w.Code)
+	}
+
+	var seeds []SeedURL
+	if err := json.Unmarshal(w.Body.Bytes(), &seeds); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	// Should have default seeds
+	if len(seeds) == 0 {
+		t.Error("Expected seeds to be returned")
+	}
+}
+
+func TestAdminCreateSeed(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	newSeed := map[string]interface{}{
+		"url":         "https://new-seed.example.com",
+		"name":        "New Seed",
+		"category":    "test",
+		"description": "Test description",
+		"trustBoost":  0.5,
+		"enabled":     true,
+	}
+
+	body, _ := json.Marshal(newSeed)
+	req, _ := http.NewRequest("POST", "/v1/admin/seeds", bytes.NewBuffer(body))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusCreated {
+		t.Errorf("Expected status 201, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var created SeedURL
+	if err := json.Unmarshal(w.Body.Bytes(), &created); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if created.ID == "" {
+		t.Error("Expected ID to be generated")
+	}
+	if created.URL != "https://new-seed.example.com" {
+		t.Errorf("Expected URL to match, got %q", created.URL)
+	}
+}
+
+func TestAdminCreateSeed_MissingURL(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	newSeed := map[string]interface{}{
+		"name":     "No URL Seed",
+		"category": "test",
+	}
+
+	body, _ := json.Marshal(newSeed)
+	req, _ := http.NewRequest("POST", "/v1/admin/seeds", bytes.NewBuffer(body))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400 for missing URL, got %d", w.Code)
+	}
+}
+
+func TestAdminUpdateSeed(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	// Create a seed first
+	newSeed := SeedURL{
+		URL:      "https://update-api-test.example.com",
+		Name:     "API Update Test",
+		Category: "test",
+	}
+	created, _ := seedStore.CreateSeed(newSeed)
+
+	// Update via API
+	updates := map[string]interface{}{
+		"name":       "Updated via API",
+		"trustBoost": 0.8,
+	}
+
+	body, _ := json.Marshal(updates)
+	req, _ := http.NewRequest("PUT", "/v1/admin/seeds/"+created.ID, bytes.NewBuffer(body))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var updated SeedURL
+	if err := json.Unmarshal(w.Body.Bytes(), &updated); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if updated.Name != "Updated via API" {
+		t.Errorf("Expected name 'Updated via API', got %q", updated.Name)
+	}
+}
+
+func TestAdminDeleteSeed(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	// Create a seed first
+	newSeed := SeedURL{
+		URL:      "https://delete-api-test.example.com",
+		Name:     "API Delete Test",
+		Category: "test",
+	}
+	created, _ := seedStore.CreateSeed(newSeed)
+
+	// Delete via API
+	req, _ := http.NewRequest("DELETE", "/v1/admin/seeds/"+created.ID, nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d", w.Code)
+	}
+
+	// Verify it's deleted
+	_, found := seedStore.GetSeed(created.ID)
+	if found {
+		t.Error("Seed should have been deleted")
+	}
+}
+
+func TestAdminDeleteSeed_NotFound(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	req, _ := http.NewRequest("DELETE", "/v1/admin/seeds/nonexistent-id", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusNotFound {
+		t.Errorf("Expected status 404, got %d", w.Code)
+	}
+}
+
+func TestAdminGetStats(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	req, _ := http.NewRequest("GET", "/v1/admin/stats", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d", w.Code)
+	}
+
+	var stats CrawlStats
+	if err := json.Unmarshal(w.Body.Bytes(), &stats); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	// Check that stats structure is populated
+	if stats.CrawlStatus == "" {
+		t.Error("Expected CrawlStatus to be set")
+	}
+	if stats.DocumentsPerCategory == nil {
+		t.Error("Expected DocumentsPerCategory to be set")
+	}
+}
+
+func TestAdminStartCrawl(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	// Reset crawl status
+	crawlStatus = "idle"
+
+	req, _ := http.NewRequest("POST", "/v1/admin/crawl/start", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusAccepted {
+		t.Errorf("Expected status 202, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var response map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if response["status"] != "started" {
+		t.Errorf("Expected status 'started', got %v", response["status"])
+	}
+}
+
+func TestAdminStartCrawl_AlreadyRunning(t *testing.T) {
+	dir := setupTestSeedStore(t)
+
+	h := &Handler{}
+	router := gin.New()
+	SetupRoutes(router, h, "test-key")
+	InitSeedStore(dir)
+
+	// Set crawl status to running
+	crawlStatus = "running"
+
+	req, _ := http.NewRequest("POST", "/v1/admin/crawl/start", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusConflict {
+		t.Errorf("Expected status 409, got %d", w.Code)
+	}
+
+	// Reset for other tests
+	crawlStatus = "idle"
+}
+
+func TestConcurrentSeedAccess(t *testing.T) {
+	setupTestSeedStore(t)
+
+	// Test concurrent reads and writes
+	done := make(chan bool, 10)
+
+	// Concurrent readers
+	for i := 0; i < 5; i++ {
+		go func() {
+			seedStore.GetAllSeeds()
+			done <- true
+		}()
+	}
+
+	// Concurrent writers
+	for i := 0; i < 5; i++ {
+		go func(n int) {
+			seed := SeedURL{
+				URL:      "https://concurrent-" + string(rune('A'+n)) + ".example.com",
+				Name:     "Concurrent Test",
+				Category: "test",
+			}
+			seedStore.CreateSeed(seed)
+			done <- true
+		}(i)
+	}
+
+	// Wait for all goroutines
+	for i := 0; i < 10; i++ {
+		<-done
+	}
+
+	// If we get here without deadlock or race, test passes
+}
--- a/edu-search-service/internal/api/handlers/orchestrator_handlers.go
+++ b/edu-search-service/internal/api/handlers/orchestrator_handlers.go
@@ -0,0 +1,207 @@
+package handlers
+
+import (
+	"net/http"
+
+	"github.com/breakpilot/edu-search-service/internal/orchestrator"
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+)
+
+// OrchestratorHandler handles orchestrator-related HTTP requests
+type OrchestratorHandler struct {
+	orchestrator *orchestrator.Orchestrator
+	repo         orchestrator.Repository
+}
+
+// NewOrchestratorHandler creates a new orchestrator handler
+func NewOrchestratorHandler(orch *orchestrator.Orchestrator, repo orchestrator.Repository) *OrchestratorHandler {
+	return &OrchestratorHandler{
+		orchestrator: orch,
+		repo:         repo,
+	}
+}
+
+// AddToQueueRequest represents a request to add a university to the crawl queue
+type AddToQueueRequest struct {
+	UniversityID string `json:"university_id" binding:"required"`
+	Priority     int    `json:"priority"`
+	InitiatedBy  string `json:"initiated_by"`
+}
+
+// GetStatus returns the current orchestrator status
+func (h *OrchestratorHandler) GetStatus(c *gin.Context) {
+	status, err := h.orchestrator.Status(c.Request.Context())
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get status", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, status)
+}
+
+// GetQueue returns all items in the crawl queue
+func (h *OrchestratorHandler) GetQueue(c *gin.Context) {
+	items, err := h.orchestrator.GetQueue(c.Request.Context())
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get queue", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"queue": items,
+		"count": len(items),
+	})
+}
+
+// AddToQueue adds a university to the crawl queue
+func (h *OrchestratorHandler) AddToQueue(c *gin.Context) {
+	var req AddToQueueRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	universityID, err := uuid.Parse(req.UniversityID)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid university_id format"})
+		return
+	}
+
+	// Default priority if not specified
+	priority := req.Priority
+	if priority == 0 {
+		priority = 5
+	}
+
+	initiatedBy := req.InitiatedBy
+	if initiatedBy == "" {
+		initiatedBy = "api"
+	}
+
+	item, err := h.orchestrator.AddUniversity(c.Request.Context(), universityID, priority, initiatedBy)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to add to queue", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusCreated, item)
+}
+
+// RemoveFromQueue removes a university from the crawl queue
+func (h *OrchestratorHandler) RemoveFromQueue(c *gin.Context) {
+	idStr := c.Param("id")
+	if idStr == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "University ID required"})
+		return
+	}
+
+	universityID, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid university_id format"})
+		return
+	}
+
+	if err := h.orchestrator.RemoveUniversity(c.Request.Context(), universityID); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to remove from queue", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{"deleted": true, "university_id": idStr})
+}
+
+// Start starts the orchestrator
+func (h *OrchestratorHandler) Start(c *gin.Context) {
+	if err := h.orchestrator.Start(); err != nil {
+		c.JSON(http.StatusConflict, gin.H{"error": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"status":  "started",
+		"message": "Orchestrator started successfully",
+	})
+}
+
+// Stop stops the orchestrator
+func (h *OrchestratorHandler) Stop(c *gin.Context) {
+	if err := h.orchestrator.Stop(); err != nil {
+		c.JSON(http.StatusConflict, gin.H{"error": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"status":  "stopped",
+		"message": "Orchestrator stopped successfully",
+	})
+}
+
+// PauseUniversity pauses crawling for a specific university
+func (h *OrchestratorHandler) PauseUniversity(c *gin.Context) {
+	idStr := c.Param("id")
+	if idStr == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "University ID required"})
+		return
+	}
+
+	universityID, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid university_id format"})
+		return
+	}
+
+	if err := h.orchestrator.PauseUniversity(c.Request.Context(), universityID); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to pause crawl", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"status":        "paused",
+		"university_id": idStr,
+	})
+}
+
+// ResumeUniversity resumes crawling for a paused university
+func (h *OrchestratorHandler) ResumeUniversity(c *gin.Context) {
+	idStr := c.Param("id")
+	if idStr == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "University ID required"})
+		return
+	}
+
+	universityID, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid university_id format"})
+		return
+	}
+
+	if err := h.orchestrator.ResumeUniversity(c.Request.Context(), universityID); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to resume crawl", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"status":        "resumed",
+		"university_id": idStr,
+	})
+}
+
+// SetupOrchestratorRoutes configures orchestrator API routes
+func SetupOrchestratorRoutes(r *gin.RouterGroup, h *OrchestratorHandler) {
+	crawl := r.Group("/crawl")
+	{
+		// Orchestrator control
+		crawl.GET("/status", h.GetStatus)
+		crawl.POST("/start", h.Start)
+		crawl.POST("/stop", h.Stop)
+
+		// Queue management
+		crawl.GET("/queue", h.GetQueue)
+		crawl.POST("/queue", h.AddToQueue)
+		crawl.DELETE("/queue/:id", h.RemoveFromQueue)
+
+		// Individual university control
+		crawl.POST("/queue/:id/pause", h.PauseUniversity)
+		crawl.POST("/queue/:id/resume", h.ResumeUniversity)
+	}
+}
--- a/edu-search-service/internal/api/handlers/orchestrator_handlers_test.go
+++ b/edu-search-service/internal/api/handlers/orchestrator_handlers_test.go
@@ -0,0 +1,659 @@
+package handlers
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/breakpilot/edu-search-service/internal/orchestrator"
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+)
+
+func init() {
+	gin.SetMode(gin.TestMode)
+}
+
+// MockRepository implements orchestrator.Repository for testing
+type MockRepository struct {
+	items        []orchestrator.CrawlQueueItem
+	failOnAdd    bool
+	failOnUpdate bool
+}
+
+func NewMockRepository() *MockRepository {
+	return &MockRepository{
+		items: make([]orchestrator.CrawlQueueItem, 0),
+	}
+}
+
+func (m *MockRepository) GetQueueItems(ctx context.Context) ([]orchestrator.CrawlQueueItem, error) {
+	return m.items, nil
+}
+
+func (m *MockRepository) GetNextInQueue(ctx context.Context) (*orchestrator.CrawlQueueItem, error) {
+	for i := range m.items {
+		if m.items[i].CurrentPhase != orchestrator.PhaseCompleted &&
+			m.items[i].CurrentPhase != orchestrator.PhaseFailed &&
+			m.items[i].CurrentPhase != orchestrator.PhasePaused {
+			return &m.items[i], nil
+		}
+	}
+	return nil, nil
+}
+
+func (m *MockRepository) AddToQueue(ctx context.Context, universityID uuid.UUID, priority int, initiatedBy string) (*orchestrator.CrawlQueueItem, error) {
+	if m.failOnAdd {
+		return nil, context.DeadlineExceeded
+	}
+
+	position := len(m.items) + 1
+	item := orchestrator.CrawlQueueItem{
+		ID:            uuid.New(),
+		UniversityID:  universityID,
+		QueuePosition: &position,
+		Priority:      priority,
+		CurrentPhase:  orchestrator.PhasePending,
+		CreatedAt:     time.Now(),
+		UpdatedAt:     time.Now(),
+	}
+	m.items = append(m.items, item)
+	return &item, nil
+}
+
+func (m *MockRepository) RemoveFromQueue(ctx context.Context, universityID uuid.UUID) error {
+	for i, item := range m.items {
+		if item.UniversityID == universityID {
+			m.items = append(m.items[:i], m.items[i+1:]...)
+			return nil
+		}
+	}
+	return nil
+}
+
+func (m *MockRepository) UpdateQueueItem(ctx context.Context, item *orchestrator.CrawlQueueItem) error {
+	if m.failOnUpdate {
+		return context.DeadlineExceeded
+	}
+	for i, existing := range m.items {
+		if existing.UniversityID == item.UniversityID {
+			m.items[i] = *item
+			return nil
+		}
+	}
+	return nil
+}
+
+func (m *MockRepository) PauseQueueItem(ctx context.Context, universityID uuid.UUID) error {
+	for i, item := range m.items {
+		if item.UniversityID == universityID {
+			m.items[i].CurrentPhase = orchestrator.PhasePaused
+			return nil
+		}
+	}
+	return nil
+}
+
+func (m *MockRepository) ResumeQueueItem(ctx context.Context, universityID uuid.UUID) error {
+	for i, item := range m.items {
+		if item.UniversityID == universityID && m.items[i].CurrentPhase == orchestrator.PhasePaused {
+			m.items[i].CurrentPhase = orchestrator.PhasePending
+			return nil
+		}
+	}
+	return nil
+}
+
+func (m *MockRepository) CompletePhase(ctx context.Context, universityID uuid.UUID, phase orchestrator.CrawlPhase, count int) error {
+	return nil
+}
+
+func (m *MockRepository) FailPhase(ctx context.Context, universityID uuid.UUID, phase orchestrator.CrawlPhase, errMsg string) error {
+	return nil
+}
+
+func (m *MockRepository) GetCompletedTodayCount(ctx context.Context) (int, error) {
+	count := 0
+	today := time.Now().Truncate(24 * time.Hour)
+	for _, item := range m.items {
+		if item.CurrentPhase == orchestrator.PhaseCompleted &&
+			item.CompletedAt != nil &&
+			item.CompletedAt.After(today) {
+			count++
+		}
+	}
+	return count, nil
+}
+
+func (m *MockRepository) GetTotalProcessedCount(ctx context.Context) (int, error) {
+	count := 0
+	for _, item := range m.items {
+		if item.CurrentPhase == orchestrator.PhaseCompleted {
+			count++
+		}
+	}
+	return count, nil
+}
+
+// MockStaffCrawler implements orchestrator.StaffCrawlerInterface
+type MockStaffCrawler struct{}
+
+func (m *MockStaffCrawler) DiscoverSampleProfessor(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	return &orchestrator.CrawlProgress{
+		Phase:      orchestrator.PhaseDiscovery,
+		ItemsFound: 1,
+	}, nil
+}
+
+func (m *MockStaffCrawler) CrawlProfessors(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	return &orchestrator.CrawlProgress{
+		Phase:      orchestrator.PhaseProfessors,
+		ItemsFound: 10,
+	}, nil
+}
+
+func (m *MockStaffCrawler) CrawlAllStaff(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	return &orchestrator.CrawlProgress{
+		Phase:      orchestrator.PhaseAllStaff,
+		ItemsFound: 50,
+	}, nil
+}
+
+// MockPubCrawler implements orchestrator.PublicationCrawlerInterface
+type MockPubCrawler struct{}
+
+func (m *MockPubCrawler) CrawlPublicationsForUniversity(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	return &orchestrator.CrawlProgress{
+		Phase:      orchestrator.PhasePublications,
+		ItemsFound: 100,
+	}, nil
+}
+
+// setupOrchestratorTestRouter creates a test router with orchestrator handler
+func setupOrchestratorTestRouter(orch *orchestrator.Orchestrator, repo orchestrator.Repository, apiKey string) *gin.Engine {
+	router := gin.New()
+
+	handler := NewOrchestratorHandler(orch, repo)
+
+	v1 := router.Group("/v1")
+	v1.Use(AuthMiddleware(apiKey))
+	SetupOrchestratorRoutes(v1, handler)
+
+	return router
+}
+
+func TestOrchestratorGetStatus(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	req, _ := http.NewRequest("GET", "/v1/crawl/status", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var status orchestrator.OrchestratorStatus
+	if err := json.Unmarshal(w.Body.Bytes(), &status); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if status.IsRunning != false {
+		t.Error("Expected orchestrator to not be running initially")
+	}
+}
+
+func TestOrchestratorGetQueue(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	req, _ := http.NewRequest("GET", "/v1/crawl/queue", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var response struct {
+		Queue []orchestrator.CrawlQueueItem `json:"queue"`
+		Count int                           `json:"count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if response.Count != 0 {
+		t.Errorf("Expected empty queue, got %d items", response.Count)
+	}
+}
+
+func TestOrchestratorAddToQueue(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	universityID := uuid.New()
+	reqBody := AddToQueueRequest{
+		UniversityID: universityID.String(),
+		Priority:     7,
+		InitiatedBy:  "test-user",
+	}
+
+	body, _ := json.Marshal(reqBody)
+	req, _ := http.NewRequest("POST", "/v1/crawl/queue", bytes.NewBuffer(body))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusCreated {
+		t.Errorf("Expected status 201, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var item orchestrator.CrawlQueueItem
+	if err := json.Unmarshal(w.Body.Bytes(), &item); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if item.UniversityID != universityID {
+		t.Errorf("Expected universityID %s, got %s", universityID, item.UniversityID)
+	}
+	if item.Priority != 7 {
+		t.Errorf("Expected priority 7, got %d", item.Priority)
+	}
+}
+
+func TestOrchestratorAddToQueue_InvalidUUID(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	reqBody := map[string]interface{}{
+		"university_id": "not-a-valid-uuid",
+		"priority":      5,
+	}
+
+	body, _ := json.Marshal(reqBody)
+	req, _ := http.NewRequest("POST", "/v1/crawl/queue", bytes.NewBuffer(body))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestOrchestratorAddToQueue_MissingUniversityID(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	reqBody := map[string]interface{}{
+		"priority": 5,
+	}
+
+	body, _ := json.Marshal(reqBody)
+	req, _ := http.NewRequest("POST", "/v1/crawl/queue", bytes.NewBuffer(body))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestOrchestratorRemoveFromQueue(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	// Add an item first
+	universityID := uuid.New()
+	repo.AddToQueue(context.Background(), universityID, 5, "test")
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	req, _ := http.NewRequest("DELETE", "/v1/crawl/queue/"+universityID.String(), nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d: %s", w.Code, w.Body.String())
+	}
+
+	// Verify it was removed
+	items, _ := repo.GetQueueItems(context.Background())
+	if len(items) != 0 {
+		t.Errorf("Expected queue to be empty, got %d items", len(items))
+	}
+}
+
+func TestOrchestratorRemoveFromQueue_InvalidUUID(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	req, _ := http.NewRequest("DELETE", "/v1/crawl/queue/invalid-uuid", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestOrchestratorStartStop(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	// Start orchestrator
+	req, _ := http.NewRequest("POST", "/v1/crawl/start", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200 on start, got %d: %s", w.Code, w.Body.String())
+	}
+
+	// Try to start again (should fail)
+	req, _ = http.NewRequest("POST", "/v1/crawl/start", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w = httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusConflict {
+		t.Errorf("Expected status 409 on duplicate start, got %d", w.Code)
+	}
+
+	// Stop orchestrator
+	req, _ = http.NewRequest("POST", "/v1/crawl/stop", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w = httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200 on stop, got %d: %s", w.Code, w.Body.String())
+	}
+
+	// Try to stop again (should fail)
+	req, _ = http.NewRequest("POST", "/v1/crawl/stop", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w = httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusConflict {
+		t.Errorf("Expected status 409 on duplicate stop, got %d", w.Code)
+	}
+}
+
+func TestOrchestratorPauseResume(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	// Add an item first
+	universityID := uuid.New()
+	repo.AddToQueue(context.Background(), universityID, 5, "test")
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	// Pause university
+	req, _ := http.NewRequest("POST", "/v1/crawl/queue/"+universityID.String()+"/pause", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200 on pause, got %d: %s", w.Code, w.Body.String())
+	}
+
+	// Verify it's paused
+	items, _ := repo.GetQueueItems(context.Background())
+	if len(items) != 1 || items[0].CurrentPhase != orchestrator.PhasePaused {
+		t.Errorf("Expected item to be paused, got phase %s", items[0].CurrentPhase)
+	}
+
+	// Resume university
+	req, _ = http.NewRequest("POST", "/v1/crawl/queue/"+universityID.String()+"/resume", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w = httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200 on resume, got %d: %s", w.Code, w.Body.String())
+	}
+
+	// Verify it's resumed
+	items, _ = repo.GetQueueItems(context.Background())
+	if len(items) != 1 || items[0].CurrentPhase == orchestrator.PhasePaused {
+		t.Errorf("Expected item to not be paused, got phase %s", items[0].CurrentPhase)
+	}
+}
+
+func TestOrchestratorPause_InvalidUUID(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	req, _ := http.NewRequest("POST", "/v1/crawl/queue/invalid-uuid/pause", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestOrchestratorNoAuth(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	// Request without auth
+	req, _ := http.NewRequest("GET", "/v1/crawl/status", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("Expected status 401, got %d", w.Code)
+	}
+}
+
+func TestOrchestratorDefaultPriority(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	// Add without priority (should default to 5)
+	universityID := uuid.New()
+	reqBody := AddToQueueRequest{
+		UniversityID: universityID.String(),
+		// Priority and InitiatedBy omitted
+	}
+
+	body, _ := json.Marshal(reqBody)
+	req, _ := http.NewRequest("POST", "/v1/crawl/queue", bytes.NewBuffer(body))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusCreated {
+		t.Errorf("Expected status 201, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var item orchestrator.CrawlQueueItem
+	if err := json.Unmarshal(w.Body.Bytes(), &item); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if item.Priority != 5 {
+		t.Errorf("Expected default priority 5, got %d", item.Priority)
+	}
+}
+
+// TestOrchestratorQueueWithNullableFields tests that queue items with NULL values
+// for optional fields (UniversityShort, LastError) are handled correctly.
+// This tests the COALESCE fix in repository.go that prevents NULL scan errors.
+func TestOrchestratorQueueWithNullableFields(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	// Add item with empty optional fields (simulates NULL from DB)
+	universityID := uuid.New()
+	item := orchestrator.CrawlQueueItem{
+		ID:              uuid.New(),
+		UniversityID:    universityID,
+		UniversityName:  "Test Universität",
+		UniversityShort: "", // Empty string (COALESCE converts NULL to '')
+		CurrentPhase:    orchestrator.PhasePending,
+		LastError:       "", // Empty string (COALESCE converts NULL to '')
+		CreatedAt:       time.Now(),
+		UpdatedAt:       time.Now(),
+	}
+	position := 1
+	item.QueuePosition = &position
+	repo.items = append(repo.items, item)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	req, _ := http.NewRequest("GET", "/v1/crawl/queue", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var response struct {
+		Queue []orchestrator.CrawlQueueItem `json:"queue"`
+		Count int                           `json:"count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if response.Count != 1 {
+		t.Errorf("Expected 1 item in queue, got %d", response.Count)
+	}
+
+	// Verify empty strings are preserved (not NULL)
+	if response.Queue[0].UniversityShort != "" {
+		t.Errorf("Expected empty UniversityShort, got %q", response.Queue[0].UniversityShort)
+	}
+	if response.Queue[0].LastError != "" {
+		t.Errorf("Expected empty LastError, got %q", response.Queue[0].LastError)
+	}
+}
+
+// TestOrchestratorQueueWithLastError tests that queue items with an error message
+// are correctly serialized and returned.
+func TestOrchestratorQueueWithLastError(t *testing.T) {
+	repo := NewMockRepository()
+	staffCrawler := &MockStaffCrawler{}
+	pubCrawler := &MockPubCrawler{}
+	orch := orchestrator.NewOrchestrator(repo, staffCrawler, pubCrawler)
+
+	// Add item with an error
+	universityID := uuid.New()
+	item := orchestrator.CrawlQueueItem{
+		ID:              uuid.New(),
+		UniversityID:    universityID,
+		UniversityName:  "Test Universität mit Fehler",
+		UniversityShort: "TUmF",
+		CurrentPhase:    orchestrator.PhaseFailed,
+		LastError:       "connection timeout after 30s",
+		RetryCount:      3,
+		MaxRetries:      3,
+		CreatedAt:       time.Now(),
+		UpdatedAt:       time.Now(),
+	}
+	position := 1
+	item.QueuePosition = &position
+	repo.items = append(repo.items, item)
+
+	router := setupOrchestratorTestRouter(orch, repo, "test-key")
+
+	req, _ := http.NewRequest("GET", "/v1/crawl/queue", nil)
+	req.Header.Set("Authorization", "Bearer test-key")
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status 200, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var response struct {
+		Queue []orchestrator.CrawlQueueItem `json:"queue"`
+		Count int                           `json:"count"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &response); err != nil {
+		t.Fatalf("Failed to parse response: %v", err)
+	}
+
+	if response.Count != 1 {
+		t.Errorf("Expected 1 item in queue, got %d", response.Count)
+	}
+
+	// Verify error message is preserved
+	if response.Queue[0].LastError != "connection timeout after 30s" {
+		t.Errorf("Expected LastError to be 'connection timeout after 30s', got %q", response.Queue[0].LastError)
+	}
+	if response.Queue[0].UniversityShort != "TUmF" {
+		t.Errorf("Expected UniversityShort 'TUmF', got %q", response.Queue[0].UniversityShort)
+	}
+}
--- a/edu-search-service/internal/api/handlers/policy_handlers.go
+++ b/edu-search-service/internal/api/handlers/policy_handlers.go
@@ -0,0 +1,700 @@
+package handlers
+
+import (
+	"net/http"
+	"time"
+
+	"github.com/breakpilot/edu-search-service/internal/policy"
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+)
+
+// PolicyHandler contains all policy-related HTTP handlers.
+type PolicyHandler struct {
+	store    *policy.Store
+	enforcer *policy.Enforcer
+}
+
+// policyHandler is the singleton instance
+var policyHandler *PolicyHandler
+
+// InitPolicyHandler initializes the policy handler with a database pool.
+func InitPolicyHandler(store *policy.Store) {
+	policyHandler = &PolicyHandler{
+		store:    store,
+		enforcer: policy.NewEnforcer(store),
+	}
+}
+
+// GetPolicyHandler returns the policy handler instance.
+func GetPolicyHandler() *PolicyHandler {
+	return policyHandler
+}
+
+// =============================================================================
+// POLICIES
+// =============================================================================
+
+// ListPolicies returns all source policies.
+func (h *PolicyHandler) ListPolicies(c *gin.Context) {
+	var filter policy.PolicyListFilter
+	if err := c.ShouldBindQuery(&filter); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid query parameters", "details": err.Error()})
+		return
+	}
+
+	// Set defaults
+	if filter.Limit <= 0 || filter.Limit > 100 {
+		filter.Limit = 50
+	}
+
+	policies, total, err := h.store.ListPolicies(c.Request.Context(), &filter)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to list policies", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"policies": policies,
+		"total":    total,
+		"limit":    filter.Limit,
+		"offset":   filter.Offset,
+	})
+}
+
+// GetPolicy returns a single policy by ID.
+func (h *PolicyHandler) GetPolicy(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid policy ID"})
+		return
+	}
+
+	p, err := h.store.GetPolicy(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get policy", "details": err.Error()})
+		return
+	}
+	if p == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Policy not found"})
+		return
+	}
+
+	c.JSON(http.StatusOK, p)
+}
+
+// CreatePolicy creates a new source policy.
+func (h *PolicyHandler) CreatePolicy(c *gin.Context) {
+	var req policy.CreateSourcePolicyRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	p, err := h.store.CreatePolicy(c.Request.Context(), &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to create policy", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionCreate, policy.AuditEntitySourcePolicy, &p.ID, nil, p, userEmail)
+
+	c.JSON(http.StatusCreated, p)
+}
+
+// UpdatePolicy updates an existing policy.
+func (h *PolicyHandler) UpdatePolicy(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid policy ID"})
+		return
+	}
+
+	// Get old value for audit
+	oldPolicy, err := h.store.GetPolicy(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get policy", "details": err.Error()})
+		return
+	}
+	if oldPolicy == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Policy not found"})
+		return
+	}
+
+	var req policy.UpdateSourcePolicyRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	p, err := h.store.UpdatePolicy(c.Request.Context(), id, &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to update policy", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionUpdate, policy.AuditEntitySourcePolicy, &p.ID, oldPolicy, p, userEmail)
+
+	c.JSON(http.StatusOK, p)
+}
+
+// =============================================================================
+// SOURCES (WHITELIST)
+// =============================================================================
+
+// ListSources returns all allowed sources.
+func (h *PolicyHandler) ListSources(c *gin.Context) {
+	var filter policy.SourceListFilter
+	if err := c.ShouldBindQuery(&filter); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid query parameters", "details": err.Error()})
+		return
+	}
+
+	// Set defaults
+	if filter.Limit <= 0 || filter.Limit > 100 {
+		filter.Limit = 50
+	}
+
+	sources, total, err := h.store.ListSources(c.Request.Context(), &filter)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to list sources", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"sources": sources,
+		"total":   total,
+		"limit":   filter.Limit,
+		"offset":  filter.Offset,
+	})
+}
+
+// GetSource returns a single source by ID.
+func (h *PolicyHandler) GetSource(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid source ID"})
+		return
+	}
+
+	source, err := h.store.GetSource(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get source", "details": err.Error()})
+		return
+	}
+	if source == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Source not found"})
+		return
+	}
+
+	c.JSON(http.StatusOK, source)
+}
+
+// CreateSource creates a new allowed source.
+func (h *PolicyHandler) CreateSource(c *gin.Context) {
+	var req policy.CreateAllowedSourceRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	source, err := h.store.CreateSource(c.Request.Context(), &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to create source", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionCreate, policy.AuditEntityAllowedSource, &source.ID, nil, source, userEmail)
+
+	c.JSON(http.StatusCreated, source)
+}
+
+// UpdateSource updates an existing source.
+func (h *PolicyHandler) UpdateSource(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid source ID"})
+		return
+	}
+
+	// Get old value for audit
+	oldSource, err := h.store.GetSource(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get source", "details": err.Error()})
+		return
+	}
+	if oldSource == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Source not found"})
+		return
+	}
+
+	var req policy.UpdateAllowedSourceRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	source, err := h.store.UpdateSource(c.Request.Context(), id, &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to update source", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionUpdate, policy.AuditEntityAllowedSource, &source.ID, oldSource, source, userEmail)
+
+	c.JSON(http.StatusOK, source)
+}
+
+// DeleteSource deletes a source.
+func (h *PolicyHandler) DeleteSource(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid source ID"})
+		return
+	}
+
+	// Get source for audit before deletion
+	source, err := h.store.GetSource(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get source", "details": err.Error()})
+		return
+	}
+	if source == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Source not found"})
+		return
+	}
+
+	if err := h.store.DeleteSource(c.Request.Context(), id); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to delete source", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionDelete, policy.AuditEntityAllowedSource, &id, source, nil, userEmail)
+
+	c.JSON(http.StatusOK, gin.H{"deleted": true, "id": id})
+}
+
+// =============================================================================
+// OPERATIONS MATRIX
+// =============================================================================
+
+// GetOperationsMatrix returns all sources with their operation permissions.
+func (h *PolicyHandler) GetOperationsMatrix(c *gin.Context) {
+	sources, err := h.store.GetOperationsMatrix(c.Request.Context())
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get operations matrix", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"sources": sources,
+		"operations": []string{
+			string(policy.OperationLookup),
+			string(policy.OperationRAG),
+			string(policy.OperationTraining),
+			string(policy.OperationExport),
+		},
+	})
+}
+
+// UpdateOperationPermission updates a single operation permission.
+func (h *PolicyHandler) UpdateOperationPermission(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid operation permission ID"})
+		return
+	}
+
+	var req policy.UpdateOperationPermissionRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	// SECURITY: Prevent enabling training
+	if req.IsAllowed != nil && *req.IsAllowed {
+		// Check if this is a training operation by querying
+		ops, _ := h.store.GetOperationsBySourceID(c.Request.Context(), id)
+		for _, op := range ops {
+			if op.ID == id && op.Operation == policy.OperationTraining {
+				c.JSON(http.StatusForbidden, gin.H{
+					"error":   "Training operations cannot be enabled",
+					"message": "Training with external data is FORBIDDEN by policy",
+				})
+				return
+			}
+		}
+	}
+
+	op, err := h.store.UpdateOperationPermission(c.Request.Context(), id, &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to update operation permission", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionUpdate, policy.AuditEntityOperationPermission, &op.ID, nil, op, userEmail)
+
+	c.JSON(http.StatusOK, op)
+}
+
+// =============================================================================
+// PII RULES
+// =============================================================================
+
+// ListPIIRules returns all PII detection rules.
+func (h *PolicyHandler) ListPIIRules(c *gin.Context) {
+	activeOnly := c.Query("active_only") == "true"
+
+	rules, err := h.store.ListPIIRules(c.Request.Context(), activeOnly)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to list PII rules", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"rules": rules,
+		"total": len(rules),
+	})
+}
+
+// GetPIIRule returns a single PII rule by ID.
+func (h *PolicyHandler) GetPIIRule(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid PII rule ID"})
+		return
+	}
+
+	rule, err := h.store.GetPIIRule(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get PII rule", "details": err.Error()})
+		return
+	}
+	if rule == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "PII rule not found"})
+		return
+	}
+
+	c.JSON(http.StatusOK, rule)
+}
+
+// CreatePIIRule creates a new PII detection rule.
+func (h *PolicyHandler) CreatePIIRule(c *gin.Context) {
+	var req policy.CreatePIIRuleRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	rule, err := h.store.CreatePIIRule(c.Request.Context(), &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to create PII rule", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionCreate, policy.AuditEntityPIIRule, &rule.ID, nil, rule, userEmail)
+
+	c.JSON(http.StatusCreated, rule)
+}
+
+// UpdatePIIRule updates an existing PII rule.
+func (h *PolicyHandler) UpdatePIIRule(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid PII rule ID"})
+		return
+	}
+
+	// Get old value for audit
+	oldRule, err := h.store.GetPIIRule(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get PII rule", "details": err.Error()})
+		return
+	}
+	if oldRule == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "PII rule not found"})
+		return
+	}
+
+	var req policy.UpdatePIIRuleRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	rule, err := h.store.UpdatePIIRule(c.Request.Context(), id, &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to update PII rule", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionUpdate, policy.AuditEntityPIIRule, &rule.ID, oldRule, rule, userEmail)
+
+	c.JSON(http.StatusOK, rule)
+}
+
+// DeletePIIRule deletes a PII rule.
+func (h *PolicyHandler) DeletePIIRule(c *gin.Context) {
+	id, err := uuid.Parse(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid PII rule ID"})
+		return
+	}
+
+	// Get rule for audit before deletion
+	rule, err := h.store.GetPIIRule(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get PII rule", "details": err.Error()})
+		return
+	}
+	if rule == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "PII rule not found"})
+		return
+	}
+
+	if err := h.store.DeletePIIRule(c.Request.Context(), id); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to delete PII rule", "details": err.Error()})
+		return
+	}
+
+	// Log audit
+	userEmail := getUserEmail(c)
+	h.enforcer.LogChange(c.Request.Context(), policy.AuditActionDelete, policy.AuditEntityPIIRule, &id, rule, nil, userEmail)
+
+	c.JSON(http.StatusOK, gin.H{"deleted": true, "id": id})
+}
+
+// TestPIIRules tests PII detection against sample text.
+func (h *PolicyHandler) TestPIIRules(c *gin.Context) {
+	var req policy.PIITestRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	response, err := h.enforcer.DetectPII(c.Request.Context(), req.Text)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to test PII detection", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, response)
+}
+
+// =============================================================================
+// AUDIT & COMPLIANCE
+// =============================================================================
+
+// ListAuditLogs returns audit log entries.
+func (h *PolicyHandler) ListAuditLogs(c *gin.Context) {
+	var filter policy.AuditLogFilter
+	if err := c.ShouldBindQuery(&filter); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid query parameters", "details": err.Error()})
+		return
+	}
+
+	// Set defaults
+	if filter.Limit <= 0 || filter.Limit > 500 {
+		filter.Limit = 100
+	}
+
+	logs, total, err := h.store.ListAuditLogs(c.Request.Context(), &filter)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to list audit logs", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"logs":   logs,
+		"total":  total,
+		"limit":  filter.Limit,
+		"offset": filter.Offset,
+	})
+}
+
+// ListBlockedContent returns blocked content log entries.
+func (h *PolicyHandler) ListBlockedContent(c *gin.Context) {
+	var filter policy.BlockedContentFilter
+	if err := c.ShouldBindQuery(&filter); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid query parameters", "details": err.Error()})
+		return
+	}
+
+	// Set defaults
+	if filter.Limit <= 0 || filter.Limit > 500 {
+		filter.Limit = 100
+	}
+
+	logs, total, err := h.store.ListBlockedContent(c.Request.Context(), &filter)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to list blocked content", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"blocked": logs,
+		"total":   total,
+		"limit":   filter.Limit,
+		"offset":  filter.Offset,
+	})
+}
+
+// CheckCompliance performs a compliance check for a URL.
+func (h *PolicyHandler) CheckCompliance(c *gin.Context) {
+	var req policy.CheckComplianceRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body", "details": err.Error()})
+		return
+	}
+
+	response, err := h.enforcer.CheckCompliance(c.Request.Context(), &req)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to check compliance", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, response)
+}
+
+// GetPolicyStats returns aggregated statistics.
+func (h *PolicyHandler) GetPolicyStats(c *gin.Context) {
+	stats, err := h.store.GetStats(c.Request.Context())
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get stats", "details": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, stats)
+}
+
+// GenerateComplianceReport generates an audit report.
+func (h *PolicyHandler) GenerateComplianceReport(c *gin.Context) {
+	var auditFilter policy.AuditLogFilter
+	var blockedFilter policy.BlockedContentFilter
+
+	// Parse date filters
+	fromStr := c.Query("from")
+	toStr := c.Query("to")
+
+	if fromStr != "" {
+		from, err := time.Parse("2006-01-02", fromStr)
+		if err == nil {
+			auditFilter.FromDate = &from
+			blockedFilter.FromDate = &from
+		}
+	}
+
+	if toStr != "" {
+		to, err := time.Parse("2006-01-02", toStr)
+		if err == nil {
+			// Add 1 day to include the end date
+			to = to.Add(24 * time.Hour)
+			auditFilter.ToDate = &to
+			blockedFilter.ToDate = &to
+		}
+	}
+
+	// No limit for report
+	auditFilter.Limit = 10000
+	blockedFilter.Limit = 10000
+
+	auditor := policy.NewAuditor(h.store)
+	report, err := auditor.GenerateAuditReport(c.Request.Context(), &auditFilter, &blockedFilter)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to generate report", "details": err.Error()})
+		return
+	}
+
+	// Set filename for download
+	format := c.Query("format")
+	if format == "download" {
+		filename := "compliance-report-" + time.Now().Format("2006-01-02") + ".json"
+		c.Header("Content-Disposition", "attachment; filename="+filename)
+		c.Header("Content-Type", "application/json")
+	}
+
+	c.JSON(http.StatusOK, report)
+}
+
+// =============================================================================
+// HELPERS
+// =============================================================================
+
+// getUserEmail extracts user email from context or headers.
+func getUserEmail(c *gin.Context) *string {
+	// Try to get from header (set by auth proxy)
+	email := c.GetHeader("X-User-Email")
+	if email != "" {
+		return &email
+	}
+
+	// Try to get from context (set by auth middleware)
+	if e, exists := c.Get("user_email"); exists {
+		if emailStr, ok := e.(string); ok {
+			return &emailStr
+		}
+	}
+
+	return nil
+}
+
+// =============================================================================
+// ROUTE SETUP
+// =============================================================================
+
+// SetupPolicyRoutes configures all policy-related routes.
+func SetupPolicyRoutes(r *gin.RouterGroup) {
+	if policyHandler == nil {
+		return
+	}
+
+	h := policyHandler
+
+	// Policies
+	r.GET("/policies", h.ListPolicies)
+	r.GET("/policies/:id", h.GetPolicy)
+	r.POST("/policies", h.CreatePolicy)
+	r.PUT("/policies/:id", h.UpdatePolicy)
+
+	// Sources (Whitelist)
+	r.GET("/sources", h.ListSources)
+	r.GET("/sources/:id", h.GetSource)
+	r.POST("/sources", h.CreateSource)
+	r.PUT("/sources/:id", h.UpdateSource)
+	r.DELETE("/sources/:id", h.DeleteSource)
+
+	// Operations Matrix
+	r.GET("/operations-matrix", h.GetOperationsMatrix)
+	r.PUT("/operations/:id", h.UpdateOperationPermission)
+
+	// PII Rules
+	r.GET("/pii-rules", h.ListPIIRules)
+	r.GET("/pii-rules/:id", h.GetPIIRule)
+	r.POST("/pii-rules", h.CreatePIIRule)
+	r.PUT("/pii-rules/:id", h.UpdatePIIRule)
+	r.DELETE("/pii-rules/:id", h.DeletePIIRule)
+	r.POST("/pii-rules/test", h.TestPIIRules)
+
+	// Audit & Compliance
+	r.GET("/policy-audit", h.ListAuditLogs)
+	r.GET("/blocked-content", h.ListBlockedContent)
+	r.POST("/check-compliance", h.CheckCompliance)
+	r.GET("/policy-stats", h.GetPolicyStats)
+	r.GET("/compliance-report", h.GenerateComplianceReport)
+}
--- a/edu-search-service/internal/api/handlers/staff_handlers.go
+++ b/edu-search-service/internal/api/handlers/staff_handlers.go
@@ -0,0 +1,374 @@
+package handlers
+
+import (
+	"fmt"
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+	"github.com/breakpilot/edu-search-service/internal/publications"
+	"github.com/breakpilot/edu-search-service/internal/staff"
+)
+
+// StaffHandlers handles staff-related API endpoints
+type StaffHandlers struct {
+	repo       *database.Repository
+	crawler    *staff.StaffCrawler
+	pubCrawler *publications.PublicationCrawler
+}
+
+// NewStaffHandlers creates new staff handlers
+func NewStaffHandlers(repo *database.Repository, email string) *StaffHandlers {
+	return &StaffHandlers{
+		repo:       repo,
+		crawler:    staff.NewStaffCrawler(repo),
+		pubCrawler: publications.NewPublicationCrawler(repo, email),
+	}
+}
+
+// SearchStaff searches for university staff
+// GET /api/v1/staff/search?q=...&university_id=...&state=...&position_type=...&is_professor=...
+func (h *StaffHandlers) SearchStaff(c *gin.Context) {
+	params := database.StaffSearchParams{
+		Query:  c.Query("q"),
+		Limit:  parseIntDefault(c.Query("limit"), 20),
+		Offset: parseIntDefault(c.Query("offset"), 0),
+	}
+
+	// Optional filters
+	if uniID := c.Query("university_id"); uniID != "" {
+		id, err := uuid.Parse(uniID)
+		if err == nil {
+			params.UniversityID = &id
+		}
+	}
+
+	if deptID := c.Query("department_id"); deptID != "" {
+		id, err := uuid.Parse(deptID)
+		if err == nil {
+			params.DepartmentID = &id
+		}
+	}
+
+	if state := c.Query("state"); state != "" {
+		params.State = &state
+	}
+
+	if uniType := c.Query("uni_type"); uniType != "" {
+		params.UniType = &uniType
+	}
+
+	if posType := c.Query("position_type"); posType != "" {
+		params.PositionType = &posType
+	}
+
+	if isProfStr := c.Query("is_professor"); isProfStr != "" {
+		isProf := isProfStr == "true" || isProfStr == "1"
+		params.IsProfessor = &isProf
+	}
+
+	result, err := h.repo.SearchStaff(c.Request.Context(), params)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, result)
+}
+
+// GetStaff gets a single staff member by ID
+// GET /api/v1/staff/:id
+func (h *StaffHandlers) GetStaff(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid staff ID"})
+		return
+	}
+
+	staff, err := h.repo.GetStaff(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Staff not found"})
+		return
+	}
+
+	c.JSON(http.StatusOK, staff)
+}
+
+// GetStaffPublications gets publications for a staff member
+// GET /api/v1/staff/:id/publications
+func (h *StaffHandlers) GetStaffPublications(c *gin.Context) {
+	idStr := c.Param("id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid staff ID"})
+		return
+	}
+
+	pubs, err := h.repo.GetStaffPublications(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"publications": pubs,
+		"total":        len(pubs),
+		"staff_id":     id,
+	})
+}
+
+// SearchPublications searches for publications
+// GET /api/v1/publications/search?q=...&year=...&pub_type=...
+func (h *StaffHandlers) SearchPublications(c *gin.Context) {
+	params := database.PublicationSearchParams{
+		Query:  c.Query("q"),
+		Limit:  parseIntDefault(c.Query("limit"), 20),
+		Offset: parseIntDefault(c.Query("offset"), 0),
+	}
+
+	if staffID := c.Query("staff_id"); staffID != "" {
+		id, err := uuid.Parse(staffID)
+		if err == nil {
+			params.StaffID = &id
+		}
+	}
+
+	if year := c.Query("year"); year != "" {
+		y := parseIntDefault(year, 0)
+		if y > 0 {
+			params.Year = &y
+		}
+	}
+
+	if yearFrom := c.Query("year_from"); yearFrom != "" {
+		y := parseIntDefault(yearFrom, 0)
+		if y > 0 {
+			params.YearFrom = &y
+		}
+	}
+
+	if yearTo := c.Query("year_to"); yearTo != "" {
+		y := parseIntDefault(yearTo, 0)
+		if y > 0 {
+			params.YearTo = &y
+		}
+	}
+
+	if pubType := c.Query("pub_type"); pubType != "" {
+		params.PubType = &pubType
+	}
+
+	result, err := h.repo.SearchPublications(c.Request.Context(), params)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, result)
+}
+
+// GetStaffStats gets statistics about staff data
+// GET /api/v1/staff/stats
+func (h *StaffHandlers) GetStaffStats(c *gin.Context) {
+	stats, err := h.repo.GetStaffStats(c.Request.Context())
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, stats)
+}
+
+// ListUniversities lists all universities
+// GET /api/v1/universities
+func (h *StaffHandlers) ListUniversities(c *gin.Context) {
+	universities, err := h.repo.ListUniversities(c.Request.Context())
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"universities": universities,
+		"total":        len(universities),
+	})
+}
+
+// StartStaffCrawl starts a staff crawl for a university
+// POST /api/v1/admin/crawl/staff
+func (h *StaffHandlers) StartStaffCrawl(c *gin.Context) {
+	var req struct {
+		UniversityID string `json:"university_id"`
+	}
+
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request"})
+		return
+	}
+
+	uniID, err := uuid.Parse(req.UniversityID)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid university ID"})
+		return
+	}
+
+	uni, err := h.repo.GetUniversity(c.Request.Context(), uniID)
+	if err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "University not found"})
+		return
+	}
+
+	// Start crawl in background
+	go func() {
+		result, err := h.crawler.CrawlUniversity(c.Request.Context(), uni)
+		if err != nil {
+			// Log error
+			return
+		}
+		_ = result
+	}()
+
+	c.JSON(http.StatusAccepted, gin.H{
+		"status":        "started",
+		"university_id": uniID,
+		"message":       "Staff crawl started in background",
+	})
+}
+
+// StartPublicationCrawl starts a publication crawl for a university
+// POST /api/v1/admin/crawl/publications
+func (h *StaffHandlers) StartPublicationCrawl(c *gin.Context) {
+	var req struct {
+		UniversityID string `json:"university_id"`
+		Limit        int    `json:"limit"`
+	}
+
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request"})
+		return
+	}
+
+	uniID, err := uuid.Parse(req.UniversityID)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid university ID"})
+		return
+	}
+
+	limit := req.Limit
+	if limit <= 0 {
+		limit = 50
+	}
+
+	// Start crawl in background
+	go func() {
+		status, err := h.pubCrawler.CrawlForUniversity(c.Request.Context(), uniID, limit)
+		if err != nil {
+			// Log error
+			return
+		}
+		_ = status
+	}()
+
+	c.JSON(http.StatusAccepted, gin.H{
+		"status":        "started",
+		"university_id": uniID,
+		"message":       "Publication crawl started in background",
+	})
+}
+
+// ResolveDOI resolves a DOI and saves the publication
+// POST /api/v1/publications/resolve-doi
+func (h *StaffHandlers) ResolveDOI(c *gin.Context) {
+	var req struct {
+		DOI     string `json:"doi"`
+		StaffID string `json:"staff_id,omitempty"`
+	}
+
+	if err := c.ShouldBindJSON(&req); err != nil || req.DOI == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "DOI is required"})
+		return
+	}
+
+	pub, err := h.pubCrawler.ResolveDOI(c.Request.Context(), req.DOI)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	// Link to staff if provided
+	if req.StaffID != "" {
+		staffID, err := uuid.Parse(req.StaffID)
+		if err == nil {
+			link := &database.StaffPublication{
+				StaffID:       staffID,
+				PublicationID: pub.ID,
+			}
+			h.repo.LinkStaffPublication(c.Request.Context(), link)
+		}
+	}
+
+	c.JSON(http.StatusOK, pub)
+}
+
+// GetCrawlStatus gets crawl status for a university
+// GET /api/v1/admin/crawl/status/:university_id
+func (h *StaffHandlers) GetCrawlStatus(c *gin.Context) {
+	idStr := c.Param("university_id")
+	id, err := uuid.Parse(idStr)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid university ID"})
+		return
+	}
+
+	status, err := h.repo.GetCrawlStatus(c.Request.Context(), id)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+
+	if status == nil {
+		c.JSON(http.StatusOK, gin.H{
+			"university_id":      id,
+			"staff_crawl_status": "never",
+			"pub_crawl_status":   "never",
+		})
+		return
+	}
+
+	c.JSON(http.StatusOK, status)
+}
+
+// Helper to parse int with default
+func parseIntDefault(s string, def int) int {
+	if s == "" {
+		return def
+	}
+	var n int
+	_, err := fmt.Sscanf(s, "%d", &n)
+	if err != nil {
+		return def
+	}
+	return n
+}
+
+// RegisterStaffRoutes registers staff-related routes
+func (h *StaffHandlers) RegisterRoutes(r *gin.RouterGroup) {
+	// Public endpoints
+	r.GET("/staff/search", h.SearchStaff)
+	r.GET("/staff/stats", h.GetStaffStats)
+	r.GET("/staff/:id", h.GetStaff)
+	r.GET("/staff/:id/publications", h.GetStaffPublications)
+
+	r.GET("/publications/search", h.SearchPublications)
+	r.POST("/publications/resolve-doi", h.ResolveDOI)
+
+	r.GET("/universities", h.ListUniversities)
+
+	// Admin endpoints
+	r.POST("/admin/crawl/staff", h.StartStaffCrawl)
+	r.POST("/admin/crawl/publications", h.StartPublicationCrawl)
+	r.GET("/admin/crawl/status/:university_id", h.GetCrawlStatus)
+}
--- a/edu-search-service/internal/config/config.go
+++ b/edu-search-service/internal/config/config.go
@@ -0,0 +1,127 @@
+package config
+
+import (
+	"os"
+	"strconv"
+)
+
+type Config struct {
+	// Server
+	Port string
+
+	// OpenSearch
+	OpenSearchURL      string
+	OpenSearchUsername string
+	OpenSearchPassword string
+	IndexName          string
+
+	// Crawler
+	UserAgent       string
+	RateLimitPerSec float64
+	MaxDepth        int
+	MaxPagesPerRun  int
+
+	// Paths
+	SeedsDir string
+	RulesDir string
+
+	// API
+	APIKey string
+
+	// Backend Integration
+	BackendURL   string // URL to Python Backend for Seeds API
+	SeedsFromAPI bool   // If true, fetch seeds from API instead of files
+
+	// Embedding/Semantic Search
+	EmbeddingProvider   string // "openai", "ollama", or "none"
+	OpenAIAPIKey        string // API Key for OpenAI embeddings
+	EmbeddingModel      string // Model name (e.g., "text-embedding-3-small")
+	EmbeddingDimension  int    // Vector dimension (1536 for OpenAI small)
+	OllamaURL           string // Ollama base URL for local embeddings
+	SemanticSearchEnabled bool  // Enable semantic search features
+
+	// Scheduler
+	SchedulerEnabled  bool   // Enable automatic crawl scheduling
+	SchedulerInterval string // Crawl interval (e.g., "24h", "168h" for weekly)
+
+	// PostgreSQL (for Staff/Publications database)
+	DBHost     string
+	DBPort     string
+	DBUser     string
+	DBPassword string
+	DBName     string
+	DBSSLMode  string
+
+	// Staff Crawler
+	StaffCrawlerEmail string // Contact email for CrossRef polite pool
+}
+
+func Load() *Config {
+	return &Config{
+		Port:               getEnv("PORT", "8084"),
+		OpenSearchURL:      getEnv("OPENSEARCH_URL", "http://opensearch:9200"),
+		OpenSearchUsername: getEnv("OPENSEARCH_USERNAME", "admin"),
+		OpenSearchPassword: getEnv("OPENSEARCH_PASSWORD", "admin"),
+		IndexName:          getEnv("INDEX_NAME", "bp_documents_v1"),
+		UserAgent:          getEnv("USER_AGENT", "BreakpilotEduCrawler/1.0 (+contact: security@breakpilot.com)"),
+		RateLimitPerSec:    getEnvFloat("RATE_LIMIT_PER_SEC", 0.2),
+		MaxDepth:           getEnvInt("MAX_DEPTH", 4),
+		MaxPagesPerRun:     getEnvInt("MAX_PAGES_PER_RUN", 500),
+		SeedsDir:           getEnv("SEEDS_DIR", "./seeds"),
+		RulesDir:           getEnv("RULES_DIR", "./rules"),
+		APIKey:             getEnv("EDU_SEARCH_API_KEY", ""),
+		BackendURL:         getEnv("BACKEND_URL", "http://backend:8000"),
+		SeedsFromAPI:       getEnvBool("SEEDS_FROM_API", true),
+		// Embedding/Semantic Search
+		EmbeddingProvider:     getEnv("EMBEDDING_PROVIDER", "none"), // "openai", "ollama", or "none"
+		OpenAIAPIKey:          getEnv("OPENAI_API_KEY", ""),
+		EmbeddingModel:        getEnv("EMBEDDING_MODEL", "text-embedding-3-small"),
+		EmbeddingDimension:    getEnvInt("EMBEDDING_DIMENSION", 1536),
+		OllamaURL:             getEnv("OLLAMA_URL", "http://ollama:11434"),
+		SemanticSearchEnabled: getEnvBool("SEMANTIC_SEARCH_ENABLED", false),
+		// Scheduler
+		SchedulerEnabled:  getEnvBool("SCHEDULER_ENABLED", false),
+		SchedulerInterval: getEnv("SCHEDULER_INTERVAL", "24h"),
+		// PostgreSQL
+		DBHost:     getEnv("DB_HOST", "postgres"),
+		DBPort:     getEnv("DB_PORT", "5432"),
+		DBUser:     getEnv("DB_USER", "postgres"),
+		DBPassword: getEnv("DB_PASSWORD", "postgres"),
+		DBName:     getEnv("DB_NAME", "breakpilot"),
+		DBSSLMode:  getEnv("DB_SSLMODE", "disable"),
+		// Staff Crawler
+		StaffCrawlerEmail: getEnv("STAFF_CRAWLER_EMAIL", "crawler@breakpilot.de"),
+	}
+}
+
+func getEnvBool(key string, fallback bool) bool {
+	if value := os.Getenv(key); value != "" {
+		return value == "true" || value == "1" || value == "yes"
+	}
+	return fallback
+}
+
+func getEnv(key, fallback string) string {
+	if value := os.Getenv(key); value != "" {
+		return value
+	}
+	return fallback
+}
+
+func getEnvInt(key string, fallback int) int {
+	if value := os.Getenv(key); value != "" {
+		if i, err := strconv.Atoi(value); err == nil {
+			return i
+		}
+	}
+	return fallback
+}
+
+func getEnvFloat(key string, fallback float64) float64 {
+	if value := os.Getenv(key); value != "" {
+		if f, err := strconv.ParseFloat(value, 64); err == nil {
+			return f
+		}
+	}
+	return fallback
+}
--- a/edu-search-service/internal/crawler/api_client.go
+++ b/edu-search-service/internal/crawler/api_client.go
@@ -0,0 +1,183 @@
+package crawler
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"time"
+)
+
+// SeedFromAPI represents a seed URL from the Backend API
+type SeedFromAPI struct {
+	URL      string  `json:"url"`
+	Trust    float64 `json:"trust"`
+	Source   string  `json:"source"`   // GOV, EDU, UNI, etc.
+	Scope    string  `json:"scope"`    // FEDERAL, STATE, etc.
+	State    string  `json:"state"`    // BW, BY, etc. (optional)
+	Depth    int     `json:"depth"`    // Crawl depth for this seed
+	Category string  `json:"category"` // Category name
+}
+
+// SeedsExportResponse represents the API response from /seeds/export/for-crawler
+type SeedsExportResponse struct {
+	Seeds      []SeedFromAPI `json:"seeds"`
+	Total      int           `json:"total"`
+	ExportedAt string        `json:"exported_at"`
+}
+
+// APIClient handles communication with the Python Backend
+type APIClient struct {
+	baseURL    string
+	httpClient *http.Client
+}
+
+// NewAPIClient creates a new API client for fetching seeds
+func NewAPIClient(backendURL string) *APIClient {
+	return &APIClient{
+		baseURL: backendURL,
+		httpClient: &http.Client{
+			Timeout: 30 * time.Second,
+		},
+	}
+}
+
+// FetchSeeds retrieves enabled seeds from the Backend API
+func (c *APIClient) FetchSeeds(ctx context.Context) (*SeedsExportResponse, error) {
+	url := fmt.Sprintf("%s/v1/edu-search/seeds/export/for-crawler", c.baseURL)
+
+	req, err := http.NewRequestWithContext(ctx, "GET", url, nil)
+	if err != nil {
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	req.Header.Set("Accept", "application/json")
+	req.Header.Set("User-Agent", "EduSearchCrawler/1.0")
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("failed to fetch seeds: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(resp.Body)
+		return nil, fmt.Errorf("API returned status %d: %s", resp.StatusCode, string(body))
+	}
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read response: %w", err)
+	}
+
+	var result SeedsExportResponse
+	if err := json.Unmarshal(body, &result); err != nil {
+		return nil, fmt.Errorf("failed to parse response: %w", err)
+	}
+
+	return &result, nil
+}
+
+// CrawlStatusReport represents a crawl status to report to the Backend
+type CrawlStatusReport struct {
+	SeedURL          string  `json:"seed_url"`
+	Status           string  `json:"status"` // "success", "error", "partial"
+	DocumentsCrawled int     `json:"documents_crawled"`
+	ErrorMessage     string  `json:"error_message,omitempty"`
+	CrawlDuration    float64 `json:"crawl_duration_seconds"`
+}
+
+// CrawlStatusResponse represents the response from crawl status endpoint
+type CrawlStatusResponse struct {
+	Success bool   `json:"success"`
+	SeedURL string `json:"seed_url"`
+	Message string `json:"message"`
+}
+
+// BulkCrawlStatusResponse represents the response from bulk crawl status endpoint
+type BulkCrawlStatusResponse struct {
+	Updated int      `json:"updated"`
+	Failed  int      `json:"failed"`
+	Errors  []string `json:"errors"`
+}
+
+// ReportStatus sends crawl status for a single seed to the Backend
+func (c *APIClient) ReportStatus(ctx context.Context, report *CrawlStatusReport) error {
+	url := fmt.Sprintf("%s/v1/edu-search/seeds/crawl-status", c.baseURL)
+
+	body, err := json.Marshal(report)
+	if err != nil {
+		return fmt.Errorf("failed to marshal report: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
+	if err != nil {
+		return fmt.Errorf("failed to create request: %w", err)
+	}
+
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Accept", "application/json")
+	req.Header.Set("User-Agent", "EduSearchCrawler/1.0")
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("failed to report status: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		respBody, _ := io.ReadAll(resp.Body)
+		return fmt.Errorf("API returned status %d: %s", resp.StatusCode, string(respBody))
+	}
+
+	return nil
+}
+
+// ReportStatusBulk sends crawl status for multiple seeds in one request
+func (c *APIClient) ReportStatusBulk(ctx context.Context, reports []*CrawlStatusReport) (*BulkCrawlStatusResponse, error) {
+	url := fmt.Sprintf("%s/v1/edu-search/seeds/crawl-status/bulk", c.baseURL)
+
+	payload := struct {
+		Updates []*CrawlStatusReport `json:"updates"`
+	}{
+		Updates: reports,
+	}
+
+	body, err := json.Marshal(payload)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal reports: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Accept", "application/json")
+	req.Header.Set("User-Agent", "EduSearchCrawler/1.0")
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("failed to report status: %w", err)
+	}
+	defer resp.Body.Close()
+
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read response: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("API returned status %d: %s", resp.StatusCode, string(respBody))
+	}
+
+	var result BulkCrawlStatusResponse
+	if err := json.Unmarshal(respBody, &result); err != nil {
+		return nil, fmt.Errorf("failed to parse response: %w", err)
+	}
+
+	return &result, nil
+}
--- a/edu-search-service/internal/crawler/api_client_test.go
+++ b/edu-search-service/internal/crawler/api_client_test.go
@@ -0,0 +1,428 @@
+package crawler
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+)
+
+func TestNewAPIClient(t *testing.T) {
+	client := NewAPIClient("http://backend:8000")
+
+	if client == nil {
+		t.Fatal("Expected non-nil client")
+	}
+
+	if client.baseURL != "http://backend:8000" {
+		t.Errorf("Expected baseURL 'http://backend:8000', got '%s'", client.baseURL)
+	}
+
+	if client.httpClient == nil {
+		t.Fatal("Expected non-nil httpClient")
+	}
+}
+
+func TestFetchSeeds_Success(t *testing.T) {
+	// Create mock server
+	mockResponse := SeedsExportResponse{
+		Seeds: []SeedFromAPI{
+			{
+				URL:      "https://www.kmk.org",
+				Trust:    0.8,
+				Source:   "GOV",
+				Scope:    "FEDERAL",
+				State:    "",
+				Depth:    3,
+				Category: "federal",
+			},
+			{
+				URL:      "https://www.km-bw.de",
+				Trust:    0.7,
+				Source:   "GOV",
+				Scope:    "STATE",
+				State:    "BW",
+				Depth:    2,
+				Category: "states",
+			},
+		},
+		Total:      2,
+		ExportedAt: "2025-01-17T10:00:00Z",
+	}
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// Verify request path
+		if r.URL.Path != "/v1/edu-search/seeds/export/for-crawler" {
+			t.Errorf("Expected path '/v1/edu-search/seeds/export/for-crawler', got '%s'", r.URL.Path)
+		}
+
+		// Verify headers
+		if r.Header.Get("Accept") != "application/json" {
+			t.Errorf("Expected Accept header 'application/json', got '%s'", r.Header.Get("Accept"))
+		}
+
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(mockResponse)
+	}))
+	defer server.Close()
+
+	// Test
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	result, err := client.FetchSeeds(ctx)
+
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+
+	if result.Total != 2 {
+		t.Errorf("Expected 2 seeds, got %d", result.Total)
+	}
+
+	if len(result.Seeds) != 2 {
+		t.Fatalf("Expected 2 seeds in array, got %d", len(result.Seeds))
+	}
+
+	// Verify first seed
+	if result.Seeds[0].URL != "https://www.kmk.org" {
+		t.Errorf("Expected URL 'https://www.kmk.org', got '%s'", result.Seeds[0].URL)
+	}
+
+	if result.Seeds[0].Trust != 0.8 {
+		t.Errorf("Expected Trust 0.8, got %f", result.Seeds[0].Trust)
+	}
+
+	if result.Seeds[0].Source != "GOV" {
+		t.Errorf("Expected Source 'GOV', got '%s'", result.Seeds[0].Source)
+	}
+
+	// Verify second seed with state
+	if result.Seeds[1].State != "BW" {
+		t.Errorf("Expected State 'BW', got '%s'", result.Seeds[1].State)
+	}
+}
+
+func TestFetchSeeds_ServerError(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+		w.Write([]byte("Internal server error"))
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	_, err := client.FetchSeeds(ctx)
+
+	if err == nil {
+		t.Fatal("Expected error for server error response")
+	}
+}
+
+func TestFetchSeeds_InvalidJSON(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		w.Write([]byte("not valid json"))
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	_, err := client.FetchSeeds(ctx)
+
+	if err == nil {
+		t.Fatal("Expected error for invalid JSON response")
+	}
+}
+
+func TestFetchSeeds_Timeout(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// Simulate slow response
+		time.Sleep(2 * time.Second)
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	// Very short timeout
+	ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
+	defer cancel()
+
+	_, err := client.FetchSeeds(ctx)
+
+	if err == nil {
+		t.Fatal("Expected timeout error")
+	}
+}
+
+func TestFetchSeeds_EmptyResponse(t *testing.T) {
+	mockResponse := SeedsExportResponse{
+		Seeds:      []SeedFromAPI{},
+		Total:      0,
+		ExportedAt: "2025-01-17T10:00:00Z",
+	}
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(mockResponse)
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	result, err := client.FetchSeeds(ctx)
+
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+
+	if result.Total != 0 {
+		t.Errorf("Expected 0 seeds, got %d", result.Total)
+	}
+
+	if len(result.Seeds) != 0 {
+		t.Errorf("Expected empty seeds array, got %d", len(result.Seeds))
+	}
+}
+
+// Tests for Crawl Status Reporting
+
+func TestReportStatus_Success(t *testing.T) {
+	var receivedReport CrawlStatusReport
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// Verify request method and path
+		if r.Method != "POST" {
+			t.Errorf("Expected POST method, got %s", r.Method)
+		}
+		if r.URL.Path != "/v1/edu-search/seeds/crawl-status" {
+			t.Errorf("Expected path '/v1/edu-search/seeds/crawl-status', got '%s'", r.URL.Path)
+		}
+		if r.Header.Get("Content-Type") != "application/json" {
+			t.Errorf("Expected Content-Type 'application/json', got '%s'", r.Header.Get("Content-Type"))
+		}
+
+		// Parse body
+		json.NewDecoder(r.Body).Decode(&receivedReport)
+
+		// Send response
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(CrawlStatusResponse{
+			Success: true,
+			SeedURL: receivedReport.SeedURL,
+			Message: "Status updated",
+		})
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	report := &CrawlStatusReport{
+		SeedURL:          "https://www.kmk.org",
+		Status:           "success",
+		DocumentsCrawled: 42,
+		CrawlDuration:    15.5,
+	}
+
+	err := client.ReportStatus(ctx, report)
+
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+
+	// Verify the report was sent correctly
+	if receivedReport.SeedURL != "https://www.kmk.org" {
+		t.Errorf("Expected SeedURL 'https://www.kmk.org', got '%s'", receivedReport.SeedURL)
+	}
+	if receivedReport.Status != "success" {
+		t.Errorf("Expected Status 'success', got '%s'", receivedReport.Status)
+	}
+	if receivedReport.DocumentsCrawled != 42 {
+		t.Errorf("Expected DocumentsCrawled 42, got %d", receivedReport.DocumentsCrawled)
+	}
+}
+
+func TestReportStatus_ServerError(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+		w.Write([]byte("Internal server error"))
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	report := &CrawlStatusReport{
+		SeedURL: "https://www.kmk.org",
+		Status:  "success",
+	}
+
+	err := client.ReportStatus(ctx, report)
+
+	if err == nil {
+		t.Fatal("Expected error for server error response")
+	}
+}
+
+func TestReportStatus_NotFound(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+		w.Write([]byte(`{"detail": "Seed nicht gefunden"}`))
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	report := &CrawlStatusReport{
+		SeedURL: "https://unknown.example.com",
+		Status:  "error",
+	}
+
+	err := client.ReportStatus(ctx, report)
+
+	if err == nil {
+		t.Fatal("Expected error for 404 response")
+	}
+}
+
+func TestReportStatusBulk_Success(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// Verify request method and path
+		if r.Method != "POST" {
+			t.Errorf("Expected POST method, got %s", r.Method)
+		}
+		if r.URL.Path != "/v1/edu-search/seeds/crawl-status/bulk" {
+			t.Errorf("Expected path '/v1/edu-search/seeds/crawl-status/bulk', got '%s'", r.URL.Path)
+		}
+
+		// Parse body
+		var payload struct {
+			Updates []*CrawlStatusReport `json:"updates"`
+		}
+		json.NewDecoder(r.Body).Decode(&payload)
+
+		// Send response
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(BulkCrawlStatusResponse{
+			Updated: len(payload.Updates),
+			Failed:  0,
+			Errors:  []string{},
+		})
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	reports := []*CrawlStatusReport{
+		{
+			SeedURL:          "https://www.kmk.org",
+			Status:           "success",
+			DocumentsCrawled: 42,
+		},
+		{
+			SeedURL:          "https://www.km-bw.de",
+			Status:           "partial",
+			DocumentsCrawled: 15,
+		},
+	}
+
+	result, err := client.ReportStatusBulk(ctx, reports)
+
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+
+	if result.Updated != 2 {
+		t.Errorf("Expected 2 updated, got %d", result.Updated)
+	}
+	if result.Failed != 0 {
+		t.Errorf("Expected 0 failed, got %d", result.Failed)
+	}
+}
+
+func TestReportStatusBulk_PartialFailure(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(BulkCrawlStatusResponse{
+			Updated: 1,
+			Failed:  1,
+			Errors:  []string{"Seed nicht gefunden: https://unknown.example.com"},
+		})
+	}))
+	defer server.Close()
+
+	client := NewAPIClient(server.URL)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	reports := []*CrawlStatusReport{
+		{SeedURL: "https://www.kmk.org", Status: "success"},
+		{SeedURL: "https://unknown.example.com", Status: "error"},
+	}
+
+	result, err := client.ReportStatusBulk(ctx, reports)
+
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+
+	if result.Updated != 1 {
+		t.Errorf("Expected 1 updated, got %d", result.Updated)
+	}
+	if result.Failed != 1 {
+		t.Errorf("Expected 1 failed, got %d", result.Failed)
+	}
+	if len(result.Errors) != 1 {
+		t.Errorf("Expected 1 error, got %d", len(result.Errors))
+	}
+}
+
+func TestCrawlStatusReport_Struct(t *testing.T) {
+	report := CrawlStatusReport{
+		SeedURL:          "https://www.example.com",
+		Status:           "success",
+		DocumentsCrawled: 100,
+		ErrorMessage:     "",
+		CrawlDuration:    25.5,
+	}
+
+	// Test JSON marshaling
+	data, err := json.Marshal(report)
+	if err != nil {
+		t.Fatalf("Failed to marshal: %v", err)
+	}
+
+	var decoded CrawlStatusReport
+	if err := json.Unmarshal(data, &decoded); err != nil {
+		t.Fatalf("Failed to unmarshal: %v", err)
+	}
+
+	if decoded.SeedURL != report.SeedURL {
+		t.Errorf("SeedURL mismatch")
+	}
+	if decoded.Status != report.Status {
+		t.Errorf("Status mismatch")
+	}
+	if decoded.DocumentsCrawled != report.DocumentsCrawled {
+		t.Errorf("DocumentsCrawled mismatch")
+	}
+	if decoded.CrawlDuration != report.CrawlDuration {
+		t.Errorf("CrawlDuration mismatch")
+	}
+}
--- a/edu-search-service/internal/crawler/crawler.go
+++ b/edu-search-service/internal/crawler/crawler.go
@@ -0,0 +1,364 @@
+package crawler
+
+import (
+	"bufio"
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"fmt"
+	"io"
+	"log"
+	"net/http"
+	"net/url"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/google/uuid"
+)
+
+// Note: API client is in the same package (api_client.go)
+
+// FetchResult contains the result of fetching a URL
+type FetchResult struct {
+	URL          string
+	CanonicalURL string
+	ContentType  string
+	StatusCode   int
+	Body         []byte
+	ContentHash  string
+	FetchTime    time.Time
+	Error        error
+}
+
+// Seed represents a URL to crawl with metadata
+type Seed struct {
+	URL        string
+	TrustBoost float64
+	Source     string // GOV, EDU, UNI, etc.
+	Scope      string // FEDERAL, STATE, etc.
+	State      string // BW, BY, etc. (optional)
+	MaxDepth   int    // Custom crawl depth for this seed
+	Category   string // Category name
+}
+
+// Crawler handles URL fetching with rate limiting and robots.txt respect
+type Crawler struct {
+	userAgent       string
+	rateLimitPerSec float64
+	maxDepth        int
+	timeout         time.Duration
+	client          *http.Client
+	denylist        map[string]bool
+	lastFetch       map[string]time.Time
+	mu              sync.Mutex
+	apiClient       *APIClient // API client for fetching seeds from Backend
+}
+
+// NewCrawler creates a new crawler instance
+func NewCrawler(userAgent string, rateLimitPerSec float64, maxDepth int) *Crawler {
+	return &Crawler{
+		userAgent:       userAgent,
+		rateLimitPerSec: rateLimitPerSec,
+		maxDepth:        maxDepth,
+		timeout:         30 * time.Second,
+		client: &http.Client{
+			Timeout: 30 * time.Second,
+			CheckRedirect: func(req *http.Request, via []*http.Request) error {
+				if len(via) >= 5 {
+					return fmt.Errorf("too many redirects")
+				}
+				return nil
+			},
+		},
+		denylist:  make(map[string]bool),
+		lastFetch: make(map[string]time.Time),
+	}
+}
+
+// SetAPIClient sets the API client for fetching seeds from Backend
+func (c *Crawler) SetAPIClient(backendURL string) {
+	c.apiClient = NewAPIClient(backendURL)
+}
+
+// LoadSeedsFromAPI fetches seeds from the Backend API
+func (c *Crawler) LoadSeedsFromAPI(ctx context.Context) ([]Seed, error) {
+	if c.apiClient == nil {
+		return nil, fmt.Errorf("API client not initialized - call SetAPIClient first")
+	}
+
+	response, err := c.apiClient.FetchSeeds(ctx)
+	if err != nil {
+		return nil, fmt.Errorf("failed to fetch seeds from API: %w", err)
+	}
+
+	seeds := make([]Seed, 0, len(response.Seeds))
+	for _, apiSeed := range response.Seeds {
+		seed := Seed{
+			URL:        apiSeed.URL,
+			TrustBoost: apiSeed.Trust,
+			Source:     apiSeed.Source,
+			Scope:      apiSeed.Scope,
+			State:      apiSeed.State,
+			MaxDepth:   apiSeed.Depth,
+			Category:   apiSeed.Category,
+		}
+		// Use default depth if not specified
+		if seed.MaxDepth <= 0 {
+			seed.MaxDepth = c.maxDepth
+		}
+		seeds = append(seeds, seed)
+	}
+
+	log.Printf("Loaded %d seeds from API (exported at: %s)", len(seeds), response.ExportedAt)
+	return seeds, nil
+}
+
+// LoadSeeds loads seed URLs from files in a directory (legacy method)
+func (c *Crawler) LoadSeeds(seedsDir string) ([]string, error) {
+	var seeds []string
+
+	files, err := filepath.Glob(filepath.Join(seedsDir, "*.txt"))
+	if err != nil {
+		return nil, err
+	}
+
+	for _, file := range files {
+		if strings.Contains(file, "denylist") {
+			// Load denylist
+			if err := c.loadDenylist(file); err != nil {
+				log.Printf("Warning: Could not load denylist %s: %v", file, err)
+			}
+			continue
+		}
+
+		fileSeeds, err := c.loadSeedFile(file)
+		if err != nil {
+			log.Printf("Warning: Could not load seed file %s: %v", file, err)
+			continue
+		}
+		seeds = append(seeds, fileSeeds...)
+	}
+
+	log.Printf("Loaded %d seeds from files, %d domains in denylist", len(seeds), len(c.denylist))
+	return seeds, nil
+}
+
+// LoadSeedsWithMetadata loads seeds from files and converts to Seed struct
+// This provides backward compatibility while allowing metadata
+func (c *Crawler) LoadSeedsWithMetadata(seedsDir string) ([]Seed, error) {
+	urlList, err := c.LoadSeeds(seedsDir)
+	if err != nil {
+		return nil, err
+	}
+
+	seeds := make([]Seed, 0, len(urlList))
+	for _, url := range urlList {
+		seeds = append(seeds, Seed{
+			URL:        url,
+			TrustBoost: 0.5, // Default trust boost
+			MaxDepth:   c.maxDepth,
+		})
+	}
+
+	return seeds, nil
+}
+
+func (c *Crawler) loadSeedFile(filename string) ([]string, error) {
+	file, err := os.Open(filename)
+	if err != nil {
+		return nil, err
+	}
+	defer file.Close()
+
+	var seeds []string
+	scanner := bufio.NewScanner(file)
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		// Skip comments and empty lines
+		if line == "" || strings.HasPrefix(line, "#") {
+			continue
+		}
+		// Extract URL (ignore comments after URL)
+		parts := strings.SplitN(line, " ", 2)
+		urlStr := strings.TrimSpace(parts[0])
+		if urlStr != "" {
+			seeds = append(seeds, urlStr)
+		}
+	}
+	return seeds, scanner.Err()
+}
+
+func (c *Crawler) loadDenylist(filename string) error {
+	file, err := os.Open(filename)
+	if err != nil {
+		return err
+	}
+	defer file.Close()
+
+	scanner := bufio.NewScanner(file)
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if line == "" || strings.HasPrefix(line, "#") {
+			continue
+		}
+		c.denylist[strings.ToLower(line)] = true
+	}
+	return scanner.Err()
+}
+
+// IsDenied checks if a domain is in the denylist
+func (c *Crawler) IsDenied(urlStr string) bool {
+	u, err := url.Parse(urlStr)
+	if err != nil {
+		return true
+	}
+
+	host := strings.ToLower(u.Host)
+
+	// Check exact match
+	if c.denylist[host] {
+		return true
+	}
+
+	// Check parent domains
+	parts := strings.Split(host, ".")
+	for i := 1; i < len(parts)-1; i++ {
+		parent := strings.Join(parts[i:], ".")
+		if c.denylist[parent] {
+			return true
+		}
+	}
+
+	return false
+}
+
+// Fetch fetches a single URL with rate limiting
+func (c *Crawler) Fetch(ctx context.Context, urlStr string) (*FetchResult, error) {
+	result := &FetchResult{
+		URL:       urlStr,
+		FetchTime: time.Now(),
+	}
+
+	// Check denylist
+	if c.IsDenied(urlStr) {
+		result.Error = fmt.Errorf("domain denied")
+		return result, result.Error
+	}
+
+	// Parse URL
+	u, err := url.Parse(urlStr)
+	if err != nil {
+		result.Error = err
+		return result, err
+	}
+
+	// Rate limiting per domain
+	c.waitForRateLimit(u.Host)
+
+	// Create request
+	req, err := http.NewRequestWithContext(ctx, "GET", urlStr, nil)
+	if err != nil {
+		result.Error = err
+		return result, err
+	}
+
+	req.Header.Set("User-Agent", c.userAgent)
+	req.Header.Set("Accept", "text/html,application/pdf,application/xhtml+xml")
+	req.Header.Set("Accept-Language", "de-DE,de;q=0.9,en;q=0.8")
+
+	// Execute request
+	resp, err := c.client.Do(req)
+	if err != nil {
+		result.Error = err
+		return result, err
+	}
+	defer resp.Body.Close()
+
+	result.StatusCode = resp.StatusCode
+	result.ContentType = resp.Header.Get("Content-Type")
+	result.CanonicalURL = resp.Request.URL.String()
+
+	if resp.StatusCode != http.StatusOK {
+		result.Error = fmt.Errorf("HTTP %d", resp.StatusCode)
+		return result, result.Error
+	}
+
+	// Read body (limit to 20MB)
+	limitedReader := io.LimitReader(resp.Body, 20*1024*1024)
+	body, err := io.ReadAll(limitedReader)
+	if err != nil {
+		result.Error = err
+		return result, err
+	}
+
+	result.Body = body
+
+	// Calculate content hash
+	hash := sha256.Sum256(body)
+	result.ContentHash = hex.EncodeToString(hash[:])
+
+	return result, nil
+}
+
+func (c *Crawler) waitForRateLimit(host string) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	minInterval := time.Duration(float64(time.Second) / c.rateLimitPerSec)
+
+	if last, ok := c.lastFetch[host]; ok {
+		elapsed := time.Since(last)
+		if elapsed < minInterval {
+			time.Sleep(minInterval - elapsed)
+		}
+	}
+
+	c.lastFetch[host] = time.Now()
+}
+
+// ExtractDomain extracts the domain from a URL
+func ExtractDomain(urlStr string) string {
+	u, err := url.Parse(urlStr)
+	if err != nil {
+		return ""
+	}
+	return u.Host
+}
+
+// GenerateDocID generates a unique document ID
+func GenerateDocID() string {
+	return uuid.New().String()
+}
+
+// NormalizeURL normalizes a URL for deduplication
+func NormalizeURL(urlStr string) string {
+	u, err := url.Parse(urlStr)
+	if err != nil {
+		return urlStr
+	}
+
+	// Remove trailing slashes
+	u.Path = strings.TrimSuffix(u.Path, "/")
+
+	// Remove common tracking parameters
+	q := u.Query()
+	for key := range q {
+		lowerKey := strings.ToLower(key)
+		if strings.HasPrefix(lowerKey, "utm_") ||
+			lowerKey == "ref" ||
+			lowerKey == "source" ||
+			lowerKey == "fbclid" ||
+			lowerKey == "gclid" {
+			q.Del(key)
+		}
+	}
+	u.RawQuery = q.Encode()
+
+	// Lowercase host
+	u.Host = strings.ToLower(u.Host)
+
+	return u.String()
+}
--- a/edu-search-service/internal/crawler/crawler_test.go
+++ b/edu-search-service/internal/crawler/crawler_test.go
@@ -0,0 +1,639 @@
+package crawler
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestNewCrawler(t *testing.T) {
+	crawler := NewCrawler("TestBot/1.0", 1.0, 3)
+
+	if crawler == nil {
+		t.Fatal("Expected non-nil crawler")
+	}
+	if crawler.userAgent != "TestBot/1.0" {
+		t.Errorf("Expected userAgent 'TestBot/1.0', got %q", crawler.userAgent)
+	}
+	if crawler.rateLimitPerSec != 1.0 {
+		t.Errorf("Expected rateLimitPerSec 1.0, got %f", crawler.rateLimitPerSec)
+	}
+	if crawler.maxDepth != 3 {
+		t.Errorf("Expected maxDepth 3, got %d", crawler.maxDepth)
+	}
+	if crawler.client == nil {
+		t.Error("Expected non-nil HTTP client")
+	}
+}
+
+func TestCrawler_LoadSeeds(t *testing.T) {
+	// Create temp directory with seed files
+	dir := t.TempDir()
+
+	// Create a seed file
+	seedContent := `# Federal education sources
+https://www.kmk.org
+https://www.bildungsserver.de
+
+# Comment line
+https://www.bpb.de # with inline comment
+`
+	if err := os.WriteFile(filepath.Join(dir, "federal.txt"), []byte(seedContent), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	// Create another seed file
+	stateContent := `https://www.km.bayern.de
+https://www.schulministerium.nrw.de
+`
+	if err := os.WriteFile(filepath.Join(dir, "states.txt"), []byte(stateContent), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	// Create denylist
+	denylistContent := `# Denylist
+facebook.com
+twitter.com
+instagram.com
+`
+	if err := os.WriteFile(filepath.Join(dir, "denylist.txt"), []byte(denylistContent), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	crawler := NewCrawler("TestBot/1.0", 1.0, 3)
+	seeds, err := crawler.LoadSeeds(dir)
+	if err != nil {
+		t.Fatalf("LoadSeeds failed: %v", err)
+	}
+
+	// Check seeds loaded
+	if len(seeds) != 5 {
+		t.Errorf("Expected 5 seeds, got %d", len(seeds))
+	}
+
+	// Check expected URLs
+	expectedURLs := []string{
+		"https://www.kmk.org",
+		"https://www.bildungsserver.de",
+		"https://www.bpb.de",
+		"https://www.km.bayern.de",
+		"https://www.schulministerium.nrw.de",
+	}
+
+	for _, expected := range expectedURLs {
+		found := false
+		for _, seed := range seeds {
+			if seed == expected {
+				found = true
+				break
+			}
+		}
+		if !found {
+			t.Errorf("Expected seed %q not found", expected)
+		}
+	}
+
+	// Check denylist loaded
+	if len(crawler.denylist) != 3 {
+		t.Errorf("Expected 3 denylist entries, got %d", len(crawler.denylist))
+	}
+}
+
+func TestCrawler_IsDenied(t *testing.T) {
+	crawler := NewCrawler("TestBot/1.0", 1.0, 3)
+	crawler.denylist = map[string]bool{
+		"facebook.com":    true,
+		"twitter.com":     true,
+		"ads.example.com": true,
+	}
+
+	tests := []struct {
+		name     string
+		url      string
+		expected bool
+	}{
+		{
+			name:     "Exact domain match",
+			url:      "https://facebook.com/page",
+			expected: true,
+		},
+		{
+			name:     "Subdomain of denied domain",
+			url:      "https://www.facebook.com/page",
+			expected: true,
+		},
+		{
+			name:     "Allowed domain",
+			url:      "https://www.kmk.org/bildung",
+			expected: false,
+		},
+		{
+			name:     "Denied subdomain",
+			url:      "https://ads.example.com/banner",
+			expected: true,
+		},
+		{
+			name:     "Parent domain allowed",
+			url:      "https://example.com/page",
+			expected: false,
+		},
+		{
+			name:     "Invalid URL scheme",
+			url:      "://invalid",
+			expected: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.IsDenied(tt.url)
+			if result != tt.expected {
+				t.Errorf("IsDenied(%q) = %v, expected %v", tt.url, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestCrawler_Fetch_Success(t *testing.T) {
+	// Create test server
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// Check user agent
+		if r.Header.Get("User-Agent") != "TestBot/1.0" {
+			t.Errorf("Expected User-Agent 'TestBot/1.0', got %q", r.Header.Get("User-Agent"))
+		}
+
+		w.Header().Set("Content-Type", "text/html; charset=utf-8")
+		w.WriteHeader(http.StatusOK)
+		w.Write([]byte("<html><body>Test content</body></html>"))
+	}))
+	defer server.Close()
+
+	crawler := NewCrawler("TestBot/1.0", 100.0, 3) // High rate limit for testing
+	ctx := context.Background()
+
+	result, err := crawler.Fetch(ctx, server.URL+"/page")
+	if err != nil {
+		t.Fatalf("Fetch failed: %v", err)
+	}
+
+	if result.StatusCode != 200 {
+		t.Errorf("Expected status 200, got %d", result.StatusCode)
+	}
+	if result.Error != nil {
+		t.Errorf("Expected no error, got %v", result.Error)
+	}
+	if !strings.Contains(result.ContentType, "text/html") {
+		t.Errorf("Expected Content-Type to contain 'text/html', got %q", result.ContentType)
+	}
+	if len(result.Body) == 0 {
+		t.Error("Expected non-empty body")
+	}
+	if result.ContentHash == "" {
+		t.Error("Expected non-empty content hash")
+	}
+	if result.FetchTime.IsZero() {
+		t.Error("Expected non-zero fetch time")
+	}
+}
+
+func TestCrawler_Fetch_DeniedDomain(t *testing.T) {
+	crawler := NewCrawler("TestBot/1.0", 100.0, 3)
+	crawler.denylist = map[string]bool{
+		"denied.com": true,
+	}
+
+	ctx := context.Background()
+	result, err := crawler.Fetch(ctx, "https://denied.com/page")
+
+	if err == nil {
+		t.Error("Expected error for denied domain")
+	}
+	if result.Error == nil {
+		t.Error("Expected error in result")
+	}
+	if !strings.Contains(result.Error.Error(), "denied") {
+		t.Errorf("Expected 'denied' in error message, got %v", result.Error)
+	}
+}
+
+func TestCrawler_Fetch_HTTPError(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	}))
+	defer server.Close()
+
+	crawler := NewCrawler("TestBot/1.0", 100.0, 3)
+	ctx := context.Background()
+
+	result, err := crawler.Fetch(ctx, server.URL+"/notfound")
+	if err == nil {
+		t.Error("Expected error for 404 response")
+	}
+	if result.StatusCode != 404 {
+		t.Errorf("Expected status 404, got %d", result.StatusCode)
+	}
+}
+
+func TestCrawler_Fetch_Redirect(t *testing.T) {
+	redirectCount := 0
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/redirect" {
+			redirectCount++
+			http.Redirect(w, r, "/final", http.StatusFound)
+			return
+		}
+		w.WriteHeader(http.StatusOK)
+		w.Write([]byte("Final content"))
+	}))
+	defer server.Close()
+
+	crawler := NewCrawler("TestBot/1.0", 100.0, 3)
+	ctx := context.Background()
+
+	result, err := crawler.Fetch(ctx, server.URL+"/redirect")
+	if err != nil {
+		t.Fatalf("Fetch failed: %v", err)
+	}
+
+	// CanonicalURL should be the final URL after redirect
+	if !strings.HasSuffix(result.CanonicalURL, "/final") {
+		t.Errorf("Expected canonical URL to end with '/final', got %q", result.CanonicalURL)
+	}
+}
+
+func TestCrawler_Fetch_Timeout(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		time.Sleep(2 * time.Second) // Delay response
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	crawler := NewCrawler("TestBot/1.0", 100.0, 3)
+	crawler.timeout = 100 * time.Millisecond // Very short timeout
+
+	ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
+	defer cancel()
+
+	_, err := crawler.Fetch(ctx, server.URL+"/slow")
+	if err == nil {
+		t.Error("Expected timeout error")
+	}
+}
+
+func TestExtractDomain(t *testing.T) {
+	tests := []struct {
+		url      string
+		expected string
+	}{
+		{
+			url:      "https://www.example.com/page",
+			expected: "www.example.com",
+		},
+		{
+			url:      "https://example.com:8080/path",
+			expected: "example.com:8080",
+		},
+		{
+			url:      "http://subdomain.example.com",
+			expected: "subdomain.example.com",
+		},
+		{
+			url:      "invalid-url",
+			expected: "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.url, func(t *testing.T) {
+			result := ExtractDomain(tt.url)
+			if result != tt.expected {
+				t.Errorf("ExtractDomain(%q) = %q, expected %q", tt.url, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestGenerateDocID(t *testing.T) {
+	id1 := GenerateDocID()
+	id2 := GenerateDocID()
+
+	if id1 == "" {
+		t.Error("Expected non-empty ID")
+	}
+	if id1 == id2 {
+		t.Error("Expected unique IDs")
+	}
+	// UUID format check (basic)
+	if len(id1) != 36 {
+		t.Errorf("Expected UUID length 36, got %d", len(id1))
+	}
+}
+
+func TestNormalizeURL(t *testing.T) {
+	tests := []struct {
+		name     string
+		url      string
+		expected string
+	}{
+		{
+			name:     "Remove trailing slash",
+			url:      "https://example.com/page/",
+			expected: "https://example.com/page",
+		},
+		{
+			name:     "Remove UTM parameters",
+			url:      "https://example.com/page?utm_source=google&utm_medium=cpc",
+			expected: "https://example.com/page",
+		},
+		{
+			name:     "Remove multiple tracking params",
+			url:      "https://example.com/page?id=123&utm_campaign=test&fbclid=abc",
+			expected: "https://example.com/page?id=123",
+		},
+		{
+			name:     "Keep non-tracking params",
+			url:      "https://example.com/search?q=test&page=2",
+			expected: "https://example.com/search?page=2&q=test",
+		},
+		{
+			name:     "Lowercase host",
+			url:      "https://EXAMPLE.COM/Page",
+			expected: "https://example.com/Page",
+		},
+		{
+			name:     "Invalid URL returns as-is",
+			url:      "not-a-url",
+			expected: "not-a-url",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := NormalizeURL(tt.url)
+			if result != tt.expected {
+				t.Errorf("NormalizeURL(%q) = %q, expected %q", tt.url, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestCrawler_RateLimit(t *testing.T) {
+	requestCount := 0
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		requestCount++
+		w.WriteHeader(http.StatusOK)
+		w.Write([]byte("OK"))
+	}))
+	defer server.Close()
+
+	// 2 requests per second = 500ms between requests
+	crawler := NewCrawler("TestBot/1.0", 2.0, 3)
+	ctx := context.Background()
+
+	start := time.Now()
+
+	// Make 3 requests
+	for i := 0; i < 3; i++ {
+		crawler.Fetch(ctx, server.URL+"/page")
+	}
+
+	elapsed := time.Since(start)
+
+	// With 2 req/sec, 3 requests should take at least 1 second (2 intervals)
+	if elapsed < 800*time.Millisecond {
+		t.Errorf("Rate limiting not working: 3 requests took only %v", elapsed)
+	}
+}
+
+func TestLoadSeedFile_EmptyLines(t *testing.T) {
+	dir := t.TempDir()
+
+	content := `
+
+https://example.com
+
+# comment
+
+https://example.org
+
+`
+	if err := os.WriteFile(filepath.Join(dir, "seeds.txt"), []byte(content), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	crawler := NewCrawler("TestBot/1.0", 1.0, 3)
+	seeds, err := crawler.LoadSeeds(dir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if len(seeds) != 2 {
+		t.Errorf("Expected 2 seeds (ignoring empty lines and comments), got %d", len(seeds))
+	}
+}
+
+func TestCrawler_Fetch_LargeBody(t *testing.T) {
+	// Create a large response (but under the limit)
+	largeBody := strings.Repeat("A", 1024*1024) // 1MB
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/plain")
+		w.WriteHeader(http.StatusOK)
+		w.Write([]byte(largeBody))
+	}))
+	defer server.Close()
+
+	crawler := NewCrawler("TestBot/1.0", 100.0, 3)
+	ctx := context.Background()
+
+	result, err := crawler.Fetch(ctx, server.URL+"/large")
+	if err != nil {
+		t.Fatalf("Fetch failed: %v", err)
+	}
+
+	if len(result.Body) != len(largeBody) {
+		t.Errorf("Expected body length %d, got %d", len(largeBody), len(result.Body))
+	}
+}
+
+// Tests for API Integration (new functionality)
+
+func TestCrawler_SetAPIClient(t *testing.T) {
+	crawler := NewCrawler("TestBot/1.0", 1.0, 3)
+
+	if crawler.apiClient != nil {
+		t.Error("Expected nil apiClient initially")
+	}
+
+	crawler.SetAPIClient("http://backend:8000")
+
+	if crawler.apiClient == nil {
+		t.Error("Expected non-nil apiClient after SetAPIClient")
+	}
+}
+
+func TestCrawler_LoadSeedsFromAPI_NotInitialized(t *testing.T) {
+	crawler := NewCrawler("TestBot/1.0", 1.0, 3)
+	ctx := context.Background()
+
+	_, err := crawler.LoadSeedsFromAPI(ctx)
+
+	if err == nil {
+		t.Error("Expected error when API client not initialized")
+	}
+}
+
+func TestCrawler_LoadSeedsFromAPI_Success(t *testing.T) {
+	// Create mock server
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		w.Write([]byte(`{
+			"seeds": [
+				{"url": "https://www.kmk.org", "trust": 0.8, "source": "GOV", "scope": "FEDERAL", "state": "", "depth": 3, "category": "federal"},
+				{"url": "https://www.km-bw.de", "trust": 0.7, "source": "GOV", "scope": "STATE", "state": "BW", "depth": 2, "category": "states"}
+			],
+			"total": 2,
+			"exported_at": "2025-01-17T10:00:00Z"
+		}`))
+	}))
+	defer server.Close()
+
+	crawler := NewCrawler("TestBot/1.0", 1.0, 4)
+	crawler.SetAPIClient(server.URL)
+	ctx := context.Background()
+
+	seeds, err := crawler.LoadSeedsFromAPI(ctx)
+
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+
+	if len(seeds) != 2 {
+		t.Fatalf("Expected 2 seeds, got %d", len(seeds))
+	}
+
+	// Check first seed
+	if seeds[0].URL != "https://www.kmk.org" {
+		t.Errorf("Expected URL 'https://www.kmk.org', got '%s'", seeds[0].URL)
+	}
+	if seeds[0].TrustBoost != 0.8 {
+		t.Errorf("Expected TrustBoost 0.8, got %f", seeds[0].TrustBoost)
+	}
+	if seeds[0].Source != "GOV" {
+		t.Errorf("Expected Source 'GOV', got '%s'", seeds[0].Source)
+	}
+	if seeds[0].MaxDepth != 3 {
+		t.Errorf("Expected MaxDepth 3, got %d", seeds[0].MaxDepth)
+	}
+
+	// Check second seed with state
+	if seeds[1].State != "BW" {
+		t.Errorf("Expected State 'BW', got '%s'", seeds[1].State)
+	}
+	if seeds[1].Category != "states" {
+		t.Errorf("Expected Category 'states', got '%s'", seeds[1].Category)
+	}
+}
+
+func TestCrawler_LoadSeedsFromAPI_DefaultDepth(t *testing.T) {
+	// Create mock server with seed that has no depth
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		w.Write([]byte(`{
+			"seeds": [
+				{"url": "https://www.example.com", "trust": 0.5, "source": "EDU", "scope": "FEDERAL", "state": "", "depth": 0, "category": "edu"}
+			],
+			"total": 1,
+			"exported_at": "2025-01-17T10:00:00Z"
+		}`))
+	}))
+	defer server.Close()
+
+	defaultDepth := 5
+	crawler := NewCrawler("TestBot/1.0", 1.0, defaultDepth)
+	crawler.SetAPIClient(server.URL)
+	ctx := context.Background()
+
+	seeds, err := crawler.LoadSeedsFromAPI(ctx)
+
+	if err != nil {
+		t.Fatalf("Unexpected error: %v", err)
+	}
+
+	// When depth is 0 or not specified, it should use crawler's default
+	if seeds[0].MaxDepth != defaultDepth {
+		t.Errorf("Expected default MaxDepth %d, got %d", defaultDepth, seeds[0].MaxDepth)
+	}
+}
+
+func TestCrawler_LoadSeedsWithMetadata(t *testing.T) {
+	dir := t.TempDir()
+
+	seedContent := `https://www.kmk.org
+https://www.bildungsserver.de`
+
+	if err := os.WriteFile(filepath.Join(dir, "seeds.txt"), []byte(seedContent), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	defaultDepth := 4
+	crawler := NewCrawler("TestBot/1.0", 1.0, defaultDepth)
+	seeds, err := crawler.LoadSeedsWithMetadata(dir)
+
+	if err != nil {
+		t.Fatalf("LoadSeedsWithMetadata failed: %v", err)
+	}
+
+	if len(seeds) != 2 {
+		t.Fatalf("Expected 2 seeds, got %d", len(seeds))
+	}
+
+	// Check default values
+	for _, seed := range seeds {
+		if seed.TrustBoost != 0.5 {
+			t.Errorf("Expected default TrustBoost 0.5, got %f", seed.TrustBoost)
+		}
+		if seed.MaxDepth != defaultDepth {
+			t.Errorf("Expected default MaxDepth %d, got %d", defaultDepth, seed.MaxDepth)
+		}
+	}
+}
+
+func TestSeed_Struct(t *testing.T) {
+	seed := Seed{
+		URL:        "https://www.example.com",
+		TrustBoost: 0.75,
+		Source:     "GOV",
+		Scope:      "STATE",
+		State:      "BY",
+		MaxDepth:   3,
+		Category:   "states",
+	}
+
+	if seed.URL != "https://www.example.com" {
+		t.Errorf("URL mismatch")
+	}
+	if seed.TrustBoost != 0.75 {
+		t.Errorf("TrustBoost mismatch")
+	}
+	if seed.Source != "GOV" {
+		t.Errorf("Source mismatch")
+	}
+	if seed.Scope != "STATE" {
+		t.Errorf("Scope mismatch")
+	}
+	if seed.State != "BY" {
+		t.Errorf("State mismatch")
+	}
+	if seed.MaxDepth != 3 {
+		t.Errorf("MaxDepth mismatch")
+	}
+	if seed.Category != "states" {
+		t.Errorf("Category mismatch")
+	}
+}
--- a/edu-search-service/internal/database/database.go
+++ b/edu-search-service/internal/database/database.go
@@ -0,0 +1,133 @@
+package database
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"os"
+	"path/filepath"
+	"time"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+)
+
+// DB holds the database connection pool
+type DB struct {
+	Pool *pgxpool.Pool
+}
+
+// Config holds database configuration
+type Config struct {
+	Host     string
+	Port     string
+	User     string
+	Password string
+	DBName   string
+	SSLMode  string
+}
+
+// NewConfig creates a new database config from environment variables
+func NewConfig() *Config {
+	return &Config{
+		Host:     getEnv("DB_HOST", "localhost"),
+		Port:     getEnv("DB_PORT", "5432"),
+		User:     getEnv("DB_USER", "postgres"),
+		Password: getEnv("DB_PASSWORD", "postgres"),
+		DBName:   getEnv("DB_NAME", "breakpilot"),
+		SSLMode:  getEnv("DB_SSLMODE", "disable"),
+	}
+}
+
+func getEnv(key, defaultValue string) string {
+	if value := os.Getenv(key); value != "" {
+		return value
+	}
+	return defaultValue
+}
+
+// ConnectionString returns the PostgreSQL connection string
+func (c *Config) ConnectionString() string {
+	return fmt.Sprintf(
+		"postgres://%s:%s@%s:%s/%s?sslmode=%s",
+		c.User, c.Password, c.Host, c.Port, c.DBName, c.SSLMode,
+	)
+}
+
+// New creates a new database connection
+func New(ctx context.Context, cfg *Config) (*DB, error) {
+	config, err := pgxpool.ParseConfig(cfg.ConnectionString())
+	if err != nil {
+		return nil, fmt.Errorf("failed to parse database config: %w", err)
+	}
+
+	// Configure connection pool
+	config.MaxConns = 10
+	config.MinConns = 2
+	config.MaxConnLifetime = time.Hour
+	config.MaxConnIdleTime = 30 * time.Minute
+
+	pool, err := pgxpool.NewWithConfig(ctx, config)
+	if err != nil {
+		return nil, fmt.Errorf("failed to create connection pool: %w", err)
+	}
+
+	// Test connection
+	if err := pool.Ping(ctx); err != nil {
+		pool.Close()
+		return nil, fmt.Errorf("failed to ping database: %w", err)
+	}
+
+	log.Printf("Connected to database %s on %s:%s", cfg.DBName, cfg.Host, cfg.Port)
+
+	return &DB{Pool: pool}, nil
+}
+
+// Close closes the database connection pool
+func (db *DB) Close() {
+	if db.Pool != nil {
+		db.Pool.Close()
+	}
+}
+
+// RunMigrations executes all SQL migrations
+func (db *DB) RunMigrations(ctx context.Context) error {
+	// Try multiple paths for migration file
+	migrationPaths := []string{
+		"migrations/001_university_staff.sql",
+		"../migrations/001_university_staff.sql",
+		"../../migrations/001_university_staff.sql",
+	}
+
+	var content []byte
+	var err error
+	var foundPath string
+
+	for _, path := range migrationPaths {
+		absPath, _ := filepath.Abs(path)
+		content, err = os.ReadFile(absPath)
+		if err == nil {
+			foundPath = absPath
+			break
+		}
+	}
+
+	if content == nil {
+		return fmt.Errorf("failed to read migration file from any path: %w", err)
+	}
+
+	log.Printf("Running migrations from: %s", foundPath)
+
+	// Execute migration
+	_, err = db.Pool.Exec(ctx, string(content))
+	if err != nil {
+		return fmt.Errorf("failed to execute migration: %w", err)
+	}
+
+	log.Println("Database migrations completed successfully")
+	return nil
+}
+
+// Health checks if the database is healthy
+func (db *DB) Health(ctx context.Context) error {
+	return db.Pool.Ping(ctx)
+}
--- a/edu-search-service/internal/database/models.go
+++ b/edu-search-service/internal/database/models.go
@@ -0,0 +1,205 @@
+package database
+
+import (
+	"time"
+
+	"github.com/google/uuid"
+)
+
+// University represents a German university/Hochschule
+type University struct {
+	ID               uuid.UUID `json:"id"`
+	Name             string    `json:"name"`
+	ShortName        *string   `json:"short_name,omitempty"`
+	URL              string    `json:"url"`
+	State            *string   `json:"state,omitempty"`
+	UniType          *string   `json:"uni_type,omitempty"`
+	StaffPagePattern *string   `json:"staff_page_pattern,omitempty"`
+	CreatedAt        time.Time `json:"created_at"`
+	UpdatedAt        time.Time `json:"updated_at"`
+}
+
+// Department represents a faculty/department at a university
+type Department struct {
+	ID           uuid.UUID  `json:"id"`
+	UniversityID uuid.UUID  `json:"university_id"`
+	Name         string     `json:"name"`
+	NameEN       *string    `json:"name_en,omitempty"`
+	URL          *string    `json:"url,omitempty"`
+	Category     *string    `json:"category,omitempty"`
+	ParentID     *uuid.UUID `json:"parent_id,omitempty"`
+	CreatedAt    time.Time  `json:"created_at"`
+	UpdatedAt    time.Time  `json:"updated_at"`
+}
+
+// UniversityStaff represents a staff member at a university
+type UniversityStaff struct {
+	ID              uuid.UUID  `json:"id"`
+	UniversityID    uuid.UUID  `json:"university_id"`
+	DepartmentID    *uuid.UUID `json:"department_id,omitempty"`
+	FirstName       *string    `json:"first_name,omitempty"`
+	LastName        string     `json:"last_name"`
+	FullName        *string    `json:"full_name,omitempty"`
+	Title           *string    `json:"title,omitempty"`
+	AcademicTitle   *string    `json:"academic_title,omitempty"`
+	Position        *string    `json:"position,omitempty"`
+	PositionType    *string    `json:"position_type,omitempty"`
+	IsProfessor     bool       `json:"is_professor"`
+	Email           *string    `json:"email,omitempty"`
+	Phone           *string    `json:"phone,omitempty"`
+	Office          *string    `json:"office,omitempty"`
+	ProfileURL      *string    `json:"profile_url,omitempty"`
+	PhotoURL        *string    `json:"photo_url,omitempty"`
+	ORCID           *string    `json:"orcid,omitempty"`
+	GoogleScholarID *string    `json:"google_scholar_id,omitempty"`
+	ResearchgateURL *string    `json:"researchgate_url,omitempty"`
+	LinkedInURL     *string    `json:"linkedin_url,omitempty"`
+	PersonalWebsite *string    `json:"personal_website,omitempty"`
+	ResearchInterests []string `json:"research_interests,omitempty"`
+	ResearchSummary *string    `json:"research_summary,omitempty"`
+	SupervisorID    *uuid.UUID `json:"supervisor_id,omitempty"`
+	TeamRole        *string    `json:"team_role,omitempty"` // leitung, mitarbeiter, sekretariat, hiwi, doktorand
+	CrawledAt       time.Time  `json:"crawled_at"`
+	LastVerified    *time.Time `json:"last_verified,omitempty"`
+	IsActive        bool       `json:"is_active"`
+	SourceURL       *string    `json:"source_url,omitempty"`
+	CreatedAt       time.Time  `json:"created_at"`
+	UpdatedAt       time.Time  `json:"updated_at"`
+
+	// Joined fields (from views)
+	UniversityName   *string `json:"university_name,omitempty"`
+	UniversityShort  *string `json:"university_short,omitempty"`
+	DepartmentName   *string `json:"department_name,omitempty"`
+	PublicationCount int     `json:"publication_count,omitempty"`
+	SupervisorName   *string `json:"supervisor_name,omitempty"`
+}
+
+// Publication represents an academic publication
+type Publication struct {
+	ID            uuid.UUID  `json:"id"`
+	Title         string     `json:"title"`
+	TitleEN       *string    `json:"title_en,omitempty"`
+	Abstract      *string    `json:"abstract,omitempty"`
+	AbstractEN    *string    `json:"abstract_en,omitempty"`
+	Year          *int       `json:"year,omitempty"`
+	Month         *int       `json:"month,omitempty"`
+	PubType       *string    `json:"pub_type,omitempty"`
+	Venue         *string    `json:"venue,omitempty"`
+	VenueShort    *string    `json:"venue_short,omitempty"`
+	Publisher     *string    `json:"publisher,omitempty"`
+	DOI           *string    `json:"doi,omitempty"`
+	ISBN          *string    `json:"isbn,omitempty"`
+	ISSN          *string    `json:"issn,omitempty"`
+	ArxivID       *string    `json:"arxiv_id,omitempty"`
+	PubmedID      *string    `json:"pubmed_id,omitempty"`
+	URL           *string    `json:"url,omitempty"`
+	PDFURL        *string    `json:"pdf_url,omitempty"`
+	CitationCount int        `json:"citation_count"`
+	Keywords      []string   `json:"keywords,omitempty"`
+	Topics        []string   `json:"topics,omitempty"`
+	Source        *string    `json:"source,omitempty"`
+	RawData       []byte     `json:"raw_data,omitempty"`
+	CrawledAt     time.Time  `json:"crawled_at"`
+	CreatedAt     time.Time  `json:"created_at"`
+	UpdatedAt     time.Time  `json:"updated_at"`
+
+	// Joined fields
+	Authors      []string `json:"authors,omitempty"`
+	AuthorCount  int      `json:"author_count,omitempty"`
+}
+
+// StaffPublication represents the N:M relationship between staff and publications
+type StaffPublication struct {
+	StaffID         uuid.UUID `json:"staff_id"`
+	PublicationID   uuid.UUID `json:"publication_id"`
+	AuthorPosition  *int      `json:"author_position,omitempty"`
+	IsCorresponding bool      `json:"is_corresponding"`
+	CreatedAt       time.Time `json:"created_at"`
+}
+
+// UniversityCrawlStatus tracks crawl progress for a university
+type UniversityCrawlStatus struct {
+	UniversityID       uuid.UUID  `json:"university_id"`
+	LastStaffCrawl     *time.Time `json:"last_staff_crawl,omitempty"`
+	StaffCrawlStatus   string     `json:"staff_crawl_status"`
+	StaffCount         int        `json:"staff_count"`
+	StaffErrors        []string   `json:"staff_errors,omitempty"`
+	LastPubCrawl       *time.Time `json:"last_pub_crawl,omitempty"`
+	PubCrawlStatus     string     `json:"pub_crawl_status"`
+	PubCount           int        `json:"pub_count"`
+	PubErrors          []string   `json:"pub_errors,omitempty"`
+	NextScheduledCrawl *time.Time `json:"next_scheduled_crawl,omitempty"`
+	CrawlPriority      int        `json:"crawl_priority"`
+	CreatedAt          time.Time  `json:"created_at"`
+	UpdatedAt          time.Time  `json:"updated_at"`
+}
+
+// CrawlHistory represents a crawl audit log entry
+type CrawlHistory struct {
+	ID           uuid.UUID  `json:"id"`
+	UniversityID *uuid.UUID `json:"university_id,omitempty"`
+	CrawlType    string     `json:"crawl_type"`
+	Status       string     `json:"status"`
+	StartedAt    time.Time  `json:"started_at"`
+	CompletedAt  *time.Time `json:"completed_at,omitempty"`
+	ItemsFound   int        `json:"items_found"`
+	ItemsNew     int        `json:"items_new"`
+	ItemsUpdated int        `json:"items_updated"`
+	Errors       []byte     `json:"errors,omitempty"`
+	Metadata     []byte     `json:"metadata,omitempty"`
+}
+
+// StaffSearchParams contains parameters for searching staff
+type StaffSearchParams struct {
+	Query        string     `json:"query,omitempty"`
+	UniversityID *uuid.UUID `json:"university_id,omitempty"`
+	DepartmentID *uuid.UUID `json:"department_id,omitempty"`
+	State        *string    `json:"state,omitempty"`
+	UniType      *string    `json:"uni_type,omitempty"`
+	PositionType *string    `json:"position_type,omitempty"`
+	IsProfessor  *bool      `json:"is_professor,omitempty"`
+	Limit        int        `json:"limit,omitempty"`
+	Offset       int        `json:"offset,omitempty"`
+}
+
+// StaffSearchResult contains search results for staff
+type StaffSearchResult struct {
+	Staff      []UniversityStaff `json:"staff"`
+	Total      int               `json:"total"`
+	Limit      int               `json:"limit"`
+	Offset     int               `json:"offset"`
+	Query      string            `json:"query,omitempty"`
+}
+
+// PublicationSearchParams contains parameters for searching publications
+type PublicationSearchParams struct {
+	Query     string     `json:"query,omitempty"`
+	StaffID   *uuid.UUID `json:"staff_id,omitempty"`
+	Year      *int       `json:"year,omitempty"`
+	YearFrom  *int       `json:"year_from,omitempty"`
+	YearTo    *int       `json:"year_to,omitempty"`
+	PubType   *string    `json:"pub_type,omitempty"`
+	Limit     int        `json:"limit,omitempty"`
+	Offset    int        `json:"offset,omitempty"`
+}
+
+// PublicationSearchResult contains search results for publications
+type PublicationSearchResult struct {
+	Publications []Publication `json:"publications"`
+	Total        int           `json:"total"`
+	Limit        int           `json:"limit"`
+	Offset       int           `json:"offset"`
+	Query        string        `json:"query,omitempty"`
+}
+
+// StaffStats contains statistics about staff data
+type StaffStats struct {
+	TotalStaff        int            `json:"total_staff"`
+	TotalProfessors   int            `json:"total_professors"`
+	TotalPublications int            `json:"total_publications"`
+	TotalUniversities int            `json:"total_universities"`
+	ByState           map[string]int `json:"by_state,omitempty"`
+	ByUniType         map[string]int `json:"by_uni_type,omitempty"`
+	ByPositionType    map[string]int `json:"by_position_type,omitempty"`
+	RecentCrawls      []CrawlHistory `json:"recent_crawls,omitempty"`
+}
--- a/edu-search-service/internal/database/repository.go
+++ b/edu-search-service/internal/database/repository.go
@@ -0,0 +1,684 @@
+package database
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"github.com/google/uuid"
+	"github.com/jackc/pgx/v5"
+)
+
+// Repository provides database operations for staff and publications
+type Repository struct {
+	db *DB
+}
+
+// NewRepository creates a new repository
+func NewRepository(db *DB) *Repository {
+	return &Repository{db: db}
+}
+
+// ============================================================================
+// UNIVERSITIES
+// ============================================================================
+
+// CreateUniversity creates a new university
+func (r *Repository) CreateUniversity(ctx context.Context, u *University) error {
+	query := `
+		INSERT INTO universities (name, short_name, url, state, uni_type, staff_page_pattern)
+		VALUES ($1, $2, $3, $4, $5, $6)
+		ON CONFLICT (url) DO UPDATE SET
+			name = EXCLUDED.name,
+			short_name = EXCLUDED.short_name,
+			state = EXCLUDED.state,
+			uni_type = EXCLUDED.uni_type,
+			staff_page_pattern = EXCLUDED.staff_page_pattern,
+			updated_at = NOW()
+		RETURNING id, created_at, updated_at
+	`
+	return r.db.Pool.QueryRow(ctx, query,
+		u.Name, u.ShortName, u.URL, u.State, u.UniType, u.StaffPagePattern,
+	).Scan(&u.ID, &u.CreatedAt, &u.UpdatedAt)
+}
+
+// GetUniversity retrieves a university by ID
+func (r *Repository) GetUniversity(ctx context.Context, id uuid.UUID) (*University, error) {
+	query := `SELECT id, name, short_name, url, state, uni_type, staff_page_pattern, created_at, updated_at
+		FROM universities WHERE id = $1`
+
+	u := &University{}
+	err := r.db.Pool.QueryRow(ctx, query, id).Scan(
+		&u.ID, &u.Name, &u.ShortName, &u.URL, &u.State, &u.UniType,
+		&u.StaffPagePattern, &u.CreatedAt, &u.UpdatedAt,
+	)
+	if err == pgx.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+	return u, nil
+}
+
+// GetUniversityByID is an alias for GetUniversity (for interface compatibility)
+func (r *Repository) GetUniversityByID(ctx context.Context, id uuid.UUID) (*University, error) {
+	return r.GetUniversity(ctx, id)
+}
+
+// GetUniversityByURL retrieves a university by URL
+func (r *Repository) GetUniversityByURL(ctx context.Context, url string) (*University, error) {
+	query := `SELECT id, name, short_name, url, state, uni_type, staff_page_pattern, created_at, updated_at
+		FROM universities WHERE url = $1`
+
+	u := &University{}
+	err := r.db.Pool.QueryRow(ctx, query, url).Scan(
+		&u.ID, &u.Name, &u.ShortName, &u.URL, &u.State, &u.UniType,
+		&u.StaffPagePattern, &u.CreatedAt, &u.UpdatedAt,
+	)
+	if err != nil {
+		return nil, err
+	}
+	return u, nil
+}
+
+// ListUniversities lists all universities
+func (r *Repository) ListUniversities(ctx context.Context) ([]University, error) {
+	query := `SELECT id, name, short_name, url, state, uni_type, staff_page_pattern, created_at, updated_at
+		FROM universities ORDER BY name`
+
+	rows, err := r.db.Pool.Query(ctx, query)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var universities []University
+	for rows.Next() {
+		var u University
+		if err := rows.Scan(
+			&u.ID, &u.Name, &u.ShortName, &u.URL, &u.State, &u.UniType,
+			&u.StaffPagePattern, &u.CreatedAt, &u.UpdatedAt,
+		); err != nil {
+			return nil, err
+		}
+		universities = append(universities, u)
+	}
+	return universities, rows.Err()
+}
+
+// ============================================================================
+// DEPARTMENTS
+// ============================================================================
+
+// CreateDepartment creates or updates a department
+func (r *Repository) CreateDepartment(ctx context.Context, d *Department) error {
+	query := `
+		INSERT INTO departments (university_id, name, name_en, url, category, parent_id)
+		VALUES ($1, $2, $3, $4, $5, $6)
+		ON CONFLICT (university_id, name) DO UPDATE SET
+			name_en = EXCLUDED.name_en,
+			url = EXCLUDED.url,
+			category = EXCLUDED.category,
+			parent_id = EXCLUDED.parent_id,
+			updated_at = NOW()
+		RETURNING id, created_at, updated_at
+	`
+	return r.db.Pool.QueryRow(ctx, query,
+		d.UniversityID, d.Name, d.NameEN, d.URL, d.Category, d.ParentID,
+	).Scan(&d.ID, &d.CreatedAt, &d.UpdatedAt)
+}
+
+// GetDepartmentByName retrieves a department by university and name
+func (r *Repository) GetDepartmentByName(ctx context.Context, uniID uuid.UUID, name string) (*Department, error) {
+	query := `SELECT id, university_id, name, name_en, url, category, parent_id, created_at, updated_at
+		FROM departments WHERE university_id = $1 AND name = $2`
+
+	d := &Department{}
+	err := r.db.Pool.QueryRow(ctx, query, uniID, name).Scan(
+		&d.ID, &d.UniversityID, &d.Name, &d.NameEN, &d.URL, &d.Category,
+		&d.ParentID, &d.CreatedAt, &d.UpdatedAt,
+	)
+	if err != nil {
+		return nil, err
+	}
+	return d, nil
+}
+
+// ============================================================================
+// STAFF
+// ============================================================================
+
+// CreateStaff creates or updates a staff member
+func (r *Repository) CreateStaff(ctx context.Context, s *UniversityStaff) error {
+	query := `
+		INSERT INTO university_staff (
+			university_id, department_id, first_name, last_name, full_name,
+			title, academic_title, position, position_type, is_professor,
+			email, phone, office, profile_url, photo_url,
+			orcid, google_scholar_id, researchgate_url, linkedin_url, personal_website,
+			research_interests, research_summary, supervisor_id, team_role, source_url
+		) VALUES (
+			$1, $2, $3, $4, $5, $6, $7, $8, $9, $10,
+			$11, $12, $13, $14, $15, $16, $17, $18, $19, $20,
+			$21, $22, $23, $24, $25
+		)
+		ON CONFLICT (university_id, first_name, last_name, COALESCE(department_id, '00000000-0000-0000-0000-000000000000'::uuid))
+		DO UPDATE SET
+			full_name = EXCLUDED.full_name,
+			title = EXCLUDED.title,
+			academic_title = EXCLUDED.academic_title,
+			position = EXCLUDED.position,
+			position_type = EXCLUDED.position_type,
+			is_professor = EXCLUDED.is_professor,
+			email = COALESCE(EXCLUDED.email, university_staff.email),
+			phone = COALESCE(EXCLUDED.phone, university_staff.phone),
+			office = COALESCE(EXCLUDED.office, university_staff.office),
+			profile_url = COALESCE(EXCLUDED.profile_url, university_staff.profile_url),
+			photo_url = COALESCE(EXCLUDED.photo_url, university_staff.photo_url),
+			orcid = COALESCE(EXCLUDED.orcid, university_staff.orcid),
+			google_scholar_id = COALESCE(EXCLUDED.google_scholar_id, university_staff.google_scholar_id),
+			researchgate_url = COALESCE(EXCLUDED.researchgate_url, university_staff.researchgate_url),
+			linkedin_url = COALESCE(EXCLUDED.linkedin_url, university_staff.linkedin_url),
+			personal_website = COALESCE(EXCLUDED.personal_website, university_staff.personal_website),
+			research_interests = COALESCE(EXCLUDED.research_interests, university_staff.research_interests),
+			research_summary = COALESCE(EXCLUDED.research_summary, university_staff.research_summary),
+			supervisor_id = COALESCE(EXCLUDED.supervisor_id, university_staff.supervisor_id),
+			team_role = COALESCE(EXCLUDED.team_role, university_staff.team_role),
+			source_url = COALESCE(EXCLUDED.source_url, university_staff.source_url),
+			crawled_at = NOW(),
+			updated_at = NOW()
+		RETURNING id, crawled_at, created_at, updated_at
+	`
+	return r.db.Pool.QueryRow(ctx, query,
+		s.UniversityID, s.DepartmentID, s.FirstName, s.LastName, s.FullName,
+		s.Title, s.AcademicTitle, s.Position, s.PositionType, s.IsProfessor,
+		s.Email, s.Phone, s.Office, s.ProfileURL, s.PhotoURL,
+		s.ORCID, s.GoogleScholarID, s.ResearchgateURL, s.LinkedInURL, s.PersonalWebsite,
+		s.ResearchInterests, s.ResearchSummary, s.SupervisorID, s.TeamRole, s.SourceURL,
+	).Scan(&s.ID, &s.CrawledAt, &s.CreatedAt, &s.UpdatedAt)
+}
+
+// GetStaff retrieves a staff member by ID
+func (r *Repository) GetStaff(ctx context.Context, id uuid.UUID) (*UniversityStaff, error) {
+	query := `SELECT * FROM v_staff_full WHERE id = $1`
+
+	s := &UniversityStaff{}
+	err := r.db.Pool.QueryRow(ctx, query, id).Scan(
+		&s.ID, &s.UniversityID, &s.DepartmentID, &s.FirstName, &s.LastName, &s.FullName,
+		&s.Title, &s.AcademicTitle, &s.Position, &s.PositionType, &s.IsProfessor,
+		&s.Email, &s.Phone, &s.Office, &s.ProfileURL, &s.PhotoURL,
+		&s.ORCID, &s.GoogleScholarID, &s.ResearchgateURL, &s.LinkedInURL, &s.PersonalWebsite,
+		&s.ResearchInterests, &s.ResearchSummary, &s.CrawledAt, &s.LastVerified, &s.IsActive, &s.SourceURL,
+		&s.CreatedAt, &s.UpdatedAt, &s.UniversityName, &s.UniversityShort, nil, nil,
+		&s.DepartmentName, nil, &s.PublicationCount,
+	)
+	if err != nil {
+		return nil, err
+	}
+	return s, nil
+}
+
+// SearchStaff searches for staff members
+func (r *Repository) SearchStaff(ctx context.Context, params StaffSearchParams) (*StaffSearchResult, error) {
+	// Build query dynamically
+	var conditions []string
+	var args []interface{}
+	argNum := 1
+
+	baseQuery := `
+		SELECT s.id, s.university_id, s.department_id, s.first_name, s.last_name, s.full_name,
+			s.title, s.academic_title, s.position, s.position_type, s.is_professor,
+			s.email, s.profile_url, s.photo_url, s.orcid,
+			s.research_interests, s.crawled_at, s.is_active,
+			u.name as university_name, u.short_name as university_short, u.state as university_state,
+			d.name as department_name,
+			(SELECT COUNT(*) FROM staff_publications sp WHERE sp.staff_id = s.id) as publication_count
+		FROM university_staff s
+		JOIN universities u ON s.university_id = u.id
+		LEFT JOIN departments d ON s.department_id = d.id
+	`
+
+	if params.Query != "" {
+		conditions = append(conditions, fmt.Sprintf(
+			`(to_tsvector('german', COALESCE(s.full_name, '') || ' ' || COALESCE(s.research_summary, '')) @@ plainto_tsquery('german', $%d)
+			 OR s.full_name ILIKE '%%' || $%d || '%%'
+			 OR s.last_name ILIKE '%%' || $%d || '%%')`,
+			argNum, argNum, argNum))
+		args = append(args, params.Query)
+		argNum++
+	}
+
+	if params.UniversityID != nil {
+		conditions = append(conditions, fmt.Sprintf("s.university_id = $%d", argNum))
+		args = append(args, *params.UniversityID)
+		argNum++
+	}
+
+	if params.DepartmentID != nil {
+		conditions = append(conditions, fmt.Sprintf("s.department_id = $%d", argNum))
+		args = append(args, *params.DepartmentID)
+		argNum++
+	}
+
+	if params.State != nil {
+		conditions = append(conditions, fmt.Sprintf("u.state = $%d", argNum))
+		args = append(args, *params.State)
+		argNum++
+	}
+
+	if params.UniType != nil {
+		conditions = append(conditions, fmt.Sprintf("u.uni_type = $%d", argNum))
+		args = append(args, *params.UniType)
+		argNum++
+	}
+
+	if params.PositionType != nil {
+		conditions = append(conditions, fmt.Sprintf("s.position_type = $%d", argNum))
+		args = append(args, *params.PositionType)
+		argNum++
+	}
+
+	if params.IsProfessor != nil {
+		conditions = append(conditions, fmt.Sprintf("s.is_professor = $%d", argNum))
+		args = append(args, *params.IsProfessor)
+		argNum++
+	}
+
+	// Build WHERE clause
+	whereClause := ""
+	if len(conditions) > 0 {
+		whereClause = "WHERE " + strings.Join(conditions, " AND ")
+	}
+
+	// Count total
+	countQuery := fmt.Sprintf("SELECT COUNT(*) FROM university_staff s JOIN universities u ON s.university_id = u.id LEFT JOIN departments d ON s.department_id = d.id %s", whereClause)
+	var total int
+	if err := r.db.Pool.QueryRow(ctx, countQuery, args...).Scan(&total); err != nil {
+		return nil, err
+	}
+
+	// Apply pagination
+	limit := params.Limit
+	if limit <= 0 {
+		limit = 20
+	}
+	if limit > 100 {
+		limit = 100
+	}
+
+	offset := params.Offset
+	if offset < 0 {
+		offset = 0
+	}
+
+	// Full query with pagination
+	fullQuery := fmt.Sprintf("%s %s ORDER BY s.is_professor DESC, s.last_name ASC LIMIT %d OFFSET %d",
+		baseQuery, whereClause, limit, offset)
+
+	rows, err := r.db.Pool.Query(ctx, fullQuery, args...)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var staff []UniversityStaff
+	for rows.Next() {
+		var s UniversityStaff
+		var uniState *string
+		if err := rows.Scan(
+			&s.ID, &s.UniversityID, &s.DepartmentID, &s.FirstName, &s.LastName, &s.FullName,
+			&s.Title, &s.AcademicTitle, &s.Position, &s.PositionType, &s.IsProfessor,
+			&s.Email, &s.ProfileURL, &s.PhotoURL, &s.ORCID,
+			&s.ResearchInterests, &s.CrawledAt, &s.IsActive,
+			&s.UniversityName, &s.UniversityShort, &uniState,
+			&s.DepartmentName, &s.PublicationCount,
+		); err != nil {
+			return nil, err
+		}
+		staff = append(staff, s)
+	}
+
+	return &StaffSearchResult{
+		Staff:  staff,
+		Total:  total,
+		Limit:  limit,
+		Offset: offset,
+		Query:  params.Query,
+	}, rows.Err()
+}
+
+// ============================================================================
+// PUBLICATIONS
+// ============================================================================
+
+// CreatePublication creates or updates a publication
+func (r *Repository) CreatePublication(ctx context.Context, p *Publication) error {
+	query := `
+		INSERT INTO publications (
+			title, title_en, abstract, abstract_en, year, month,
+			pub_type, venue, venue_short, publisher,
+			doi, isbn, issn, arxiv_id, pubmed_id,
+			url, pdf_url, citation_count, keywords, topics, source, raw_data
+		) VALUES (
+			$1, $2, $3, $4, $5, $6, $7, $8, $9, $10,
+			$11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22
+		)
+		ON CONFLICT (doi) WHERE doi IS NOT NULL DO UPDATE SET
+			title = EXCLUDED.title,
+			abstract = EXCLUDED.abstract,
+			year = EXCLUDED.year,
+			venue = EXCLUDED.venue,
+			citation_count = EXCLUDED.citation_count,
+			updated_at = NOW()
+		RETURNING id, crawled_at, created_at, updated_at
+	`
+
+	// Handle potential duplicate without DOI
+	err := r.db.Pool.QueryRow(ctx, query,
+		p.Title, p.TitleEN, p.Abstract, p.AbstractEN, p.Year, p.Month,
+		p.PubType, p.Venue, p.VenueShort, p.Publisher,
+		p.DOI, p.ISBN, p.ISSN, p.ArxivID, p.PubmedID,
+		p.URL, p.PDFURL, p.CitationCount, p.Keywords, p.Topics, p.Source, p.RawData,
+	).Scan(&p.ID, &p.CrawledAt, &p.CreatedAt, &p.UpdatedAt)
+
+	if err != nil && strings.Contains(err.Error(), "duplicate") {
+		// Try to find existing publication by title and year
+		findQuery := `SELECT id FROM publications WHERE title = $1 AND year = $2`
+		err = r.db.Pool.QueryRow(ctx, findQuery, p.Title, p.Year).Scan(&p.ID)
+	}
+
+	return err
+}
+
+// LinkStaffPublication creates a link between staff and publication
+func (r *Repository) LinkStaffPublication(ctx context.Context, sp *StaffPublication) error {
+	query := `
+		INSERT INTO staff_publications (staff_id, publication_id, author_position, is_corresponding)
+		VALUES ($1, $2, $3, $4)
+		ON CONFLICT (staff_id, publication_id) DO UPDATE SET
+			author_position = EXCLUDED.author_position,
+			is_corresponding = EXCLUDED.is_corresponding
+	`
+	_, err := r.db.Pool.Exec(ctx, query,
+		sp.StaffID, sp.PublicationID, sp.AuthorPosition, sp.IsCorresponding,
+	)
+	return err
+}
+
+// GetStaffPublications retrieves all publications for a staff member
+func (r *Repository) GetStaffPublications(ctx context.Context, staffID uuid.UUID) ([]Publication, error) {
+	query := `
+		SELECT p.id, p.title, p.abstract, p.year, p.pub_type, p.venue, p.doi, p.url, p.citation_count
+		FROM publications p
+		JOIN staff_publications sp ON p.id = sp.publication_id
+		WHERE sp.staff_id = $1
+		ORDER BY p.year DESC NULLS LAST, p.title
+	`
+
+	rows, err := r.db.Pool.Query(ctx, query, staffID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var pubs []Publication
+	for rows.Next() {
+		var p Publication
+		if err := rows.Scan(
+			&p.ID, &p.Title, &p.Abstract, &p.Year, &p.PubType, &p.Venue, &p.DOI, &p.URL, &p.CitationCount,
+		); err != nil {
+			return nil, err
+		}
+		pubs = append(pubs, p)
+	}
+	return pubs, rows.Err()
+}
+
+// SearchPublications searches for publications
+func (r *Repository) SearchPublications(ctx context.Context, params PublicationSearchParams) (*PublicationSearchResult, error) {
+	var conditions []string
+	var args []interface{}
+	argNum := 1
+
+	if params.Query != "" {
+		conditions = append(conditions, fmt.Sprintf(
+			`to_tsvector('german', COALESCE(title, '') || ' ' || COALESCE(abstract, '')) @@ plainto_tsquery('german', $%d)`,
+			argNum))
+		args = append(args, params.Query)
+		argNum++
+	}
+
+	if params.StaffID != nil {
+		conditions = append(conditions, fmt.Sprintf(
+			`id IN (SELECT publication_id FROM staff_publications WHERE staff_id = $%d)`,
+			argNum))
+		args = append(args, *params.StaffID)
+		argNum++
+	}
+
+	if params.Year != nil {
+		conditions = append(conditions, fmt.Sprintf("year = $%d", argNum))
+		args = append(args, *params.Year)
+		argNum++
+	}
+
+	if params.YearFrom != nil {
+		conditions = append(conditions, fmt.Sprintf("year >= $%d", argNum))
+		args = append(args, *params.YearFrom)
+		argNum++
+	}
+
+	if params.YearTo != nil {
+		conditions = append(conditions, fmt.Sprintf("year <= $%d", argNum))
+		args = append(args, *params.YearTo)
+		argNum++
+	}
+
+	if params.PubType != nil {
+		conditions = append(conditions, fmt.Sprintf("pub_type = $%d", argNum))
+		args = append(args, *params.PubType)
+		argNum++
+	}
+
+	whereClause := ""
+	if len(conditions) > 0 {
+		whereClause = "WHERE " + strings.Join(conditions, " AND ")
+	}
+
+	// Count
+	countQuery := fmt.Sprintf("SELECT COUNT(*) FROM publications %s", whereClause)
+	var total int
+	if err := r.db.Pool.QueryRow(ctx, countQuery, args...).Scan(&total); err != nil {
+		return nil, err
+	}
+
+	// Pagination
+	limit := params.Limit
+	if limit <= 0 {
+		limit = 20
+	}
+	offset := params.Offset
+
+	// Query
+	query := fmt.Sprintf(`
+		SELECT id, title, abstract, year, pub_type, venue, doi, url, citation_count, keywords
+		FROM publications %s
+		ORDER BY year DESC NULLS LAST, citation_count DESC
+		LIMIT %d OFFSET %d
+	`, whereClause, limit, offset)
+
+	rows, err := r.db.Pool.Query(ctx, query, args...)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var pubs []Publication
+	for rows.Next() {
+		var p Publication
+		if err := rows.Scan(
+			&p.ID, &p.Title, &p.Abstract, &p.Year, &p.PubType, &p.Venue, &p.DOI, &p.URL, &p.CitationCount, &p.Keywords,
+		); err != nil {
+			return nil, err
+		}
+		pubs = append(pubs, p)
+	}
+
+	return &PublicationSearchResult{
+		Publications: pubs,
+		Total:        total,
+		Limit:        limit,
+		Offset:       offset,
+		Query:        params.Query,
+	}, rows.Err()
+}
+
+// ============================================================================
+// CRAWL STATUS
+// ============================================================================
+
+// UpdateCrawlStatus updates crawl status for a university
+func (r *Repository) UpdateCrawlStatus(ctx context.Context, status *UniversityCrawlStatus) error {
+	query := `
+		INSERT INTO university_crawl_status (
+			university_id, last_staff_crawl, staff_crawl_status, staff_count, staff_errors,
+			last_pub_crawl, pub_crawl_status, pub_count, pub_errors,
+			next_scheduled_crawl, crawl_priority
+		) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11)
+		ON CONFLICT (university_id) DO UPDATE SET
+			last_staff_crawl = EXCLUDED.last_staff_crawl,
+			staff_crawl_status = EXCLUDED.staff_crawl_status,
+			staff_count = EXCLUDED.staff_count,
+			staff_errors = EXCLUDED.staff_errors,
+			last_pub_crawl = EXCLUDED.last_pub_crawl,
+			pub_crawl_status = EXCLUDED.pub_crawl_status,
+			pub_count = EXCLUDED.pub_count,
+			pub_errors = EXCLUDED.pub_errors,
+			next_scheduled_crawl = EXCLUDED.next_scheduled_crawl,
+			crawl_priority = EXCLUDED.crawl_priority,
+			updated_at = NOW()
+	`
+	_, err := r.db.Pool.Exec(ctx, query,
+		status.UniversityID, status.LastStaffCrawl, status.StaffCrawlStatus, status.StaffCount, status.StaffErrors,
+		status.LastPubCrawl, status.PubCrawlStatus, status.PubCount, status.PubErrors,
+		status.NextScheduledCrawl, status.CrawlPriority,
+	)
+	return err
+}
+
+// GetCrawlStatus retrieves crawl status for a university
+func (r *Repository) GetCrawlStatus(ctx context.Context, uniID uuid.UUID) (*UniversityCrawlStatus, error) {
+	query := `SELECT * FROM university_crawl_status WHERE university_id = $1`
+
+	s := &UniversityCrawlStatus{}
+	err := r.db.Pool.QueryRow(ctx, query, uniID).Scan(
+		&s.UniversityID, &s.LastStaffCrawl, &s.StaffCrawlStatus, &s.StaffCount, &s.StaffErrors,
+		&s.LastPubCrawl, &s.PubCrawlStatus, &s.PubCount, &s.PubErrors,
+		&s.NextScheduledCrawl, &s.CrawlPriority, &s.CreatedAt, &s.UpdatedAt,
+	)
+	if err == pgx.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+	return s, nil
+}
+
+// ============================================================================
+// STATS
+// ============================================================================
+
+// GetStaffStats retrieves statistics about staff data
+func (r *Repository) GetStaffStats(ctx context.Context) (*StaffStats, error) {
+	stats := &StaffStats{
+		ByState:        make(map[string]int),
+		ByUniType:      make(map[string]int),
+		ByPositionType: make(map[string]int),
+	}
+
+	// Basic counts
+	queries := []struct {
+		query string
+		dest  *int
+	}{
+		{"SELECT COUNT(*) FROM university_staff WHERE is_active = true", &stats.TotalStaff},
+		{"SELECT COUNT(*) FROM university_staff WHERE is_professor = true AND is_active = true", &stats.TotalProfessors},
+		{"SELECT COUNT(*) FROM publications", &stats.TotalPublications},
+		{"SELECT COUNT(*) FROM universities", &stats.TotalUniversities},
+	}
+
+	for _, q := range queries {
+		if err := r.db.Pool.QueryRow(ctx, q.query).Scan(q.dest); err != nil {
+			return nil, err
+		}
+	}
+
+	// By state
+	rows, err := r.db.Pool.Query(ctx, `
+		SELECT COALESCE(u.state, 'unknown'), COUNT(*)
+		FROM university_staff s
+		JOIN universities u ON s.university_id = u.id
+		WHERE s.is_active = true
+		GROUP BY u.state
+	`)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	for rows.Next() {
+		var state string
+		var count int
+		if err := rows.Scan(&state, &count); err != nil {
+			return nil, err
+		}
+		stats.ByState[state] = count
+	}
+
+	// By uni type
+	rows2, err := r.db.Pool.Query(ctx, `
+		SELECT COALESCE(u.uni_type, 'unknown'), COUNT(*)
+		FROM university_staff s
+		JOIN universities u ON s.university_id = u.id
+		WHERE s.is_active = true
+		GROUP BY u.uni_type
+	`)
+	if err != nil {
+		return nil, err
+	}
+	defer rows2.Close()
+
+	for rows2.Next() {
+		var uniType string
+		var count int
+		if err := rows2.Scan(&uniType, &count); err != nil {
+			return nil, err
+		}
+		stats.ByUniType[uniType] = count
+	}
+
+	// By position type
+	rows3, err := r.db.Pool.Query(ctx, `
+		SELECT COALESCE(position_type, 'unknown'), COUNT(*)
+		FROM university_staff
+		WHERE is_active = true
+		GROUP BY position_type
+	`)
+	if err != nil {
+		return nil, err
+	}
+	defer rows3.Close()
+
+	for rows3.Next() {
+		var posType string
+		var count int
+		if err := rows3.Scan(&posType, &count); err != nil {
+			return nil, err
+		}
+		stats.ByPositionType[posType] = count
+	}
+
+	return stats, nil
+}
--- a/edu-search-service/internal/embedding/embedding.go
+++ b/edu-search-service/internal/embedding/embedding.go
@@ -0,0 +1,332 @@
+package embedding
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"time"
+)
+
+// EmbeddingProvider defines the interface for embedding services
+type EmbeddingProvider interface {
+	// Embed generates embeddings for the given text
+	Embed(ctx context.Context, text string) ([]float32, error)
+
+	// EmbedBatch generates embeddings for multiple texts
+	EmbedBatch(ctx context.Context, texts []string) ([][]float32, error)
+
+	// Dimension returns the embedding vector dimension
+	Dimension() int
+}
+
+// Service wraps an embedding provider
+type Service struct {
+	provider  EmbeddingProvider
+	dimension int
+	enabled   bool
+}
+
+// NewService creates a new embedding service based on configuration
+func NewService(provider, apiKey, model, ollamaURL string, dimension int, enabled bool) (*Service, error) {
+	if !enabled {
+		return &Service{
+			provider:  nil,
+			dimension: dimension,
+			enabled:   false,
+		}, nil
+	}
+
+	var p EmbeddingProvider
+	var err error
+
+	switch provider {
+	case "openai":
+		if apiKey == "" {
+			return nil, errors.New("OpenAI API key required for openai provider")
+		}
+		p = NewOpenAIProvider(apiKey, model, dimension)
+	case "ollama":
+		p, err = NewOllamaProvider(ollamaURL, model, dimension)
+		if err != nil {
+			return nil, err
+		}
+	case "none", "":
+		return &Service{
+			provider:  nil,
+			dimension: dimension,
+			enabled:   false,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unknown embedding provider: %s", provider)
+	}
+
+	return &Service{
+		provider:  p,
+		dimension: dimension,
+		enabled:   true,
+	}, nil
+}
+
+// IsEnabled returns true if semantic search is enabled
+func (s *Service) IsEnabled() bool {
+	return s.enabled && s.provider != nil
+}
+
+// Embed generates embedding for a single text
+func (s *Service) Embed(ctx context.Context, text string) ([]float32, error) {
+	if !s.IsEnabled() {
+		return nil, errors.New("embedding service not enabled")
+	}
+	return s.provider.Embed(ctx, text)
+}
+
+// EmbedBatch generates embeddings for multiple texts
+func (s *Service) EmbedBatch(ctx context.Context, texts []string) ([][]float32, error) {
+	if !s.IsEnabled() {
+		return nil, errors.New("embedding service not enabled")
+	}
+	return s.provider.EmbedBatch(ctx, texts)
+}
+
+// Dimension returns the configured embedding dimension
+func (s *Service) Dimension() int {
+	return s.dimension
+}
+
+// =====================================================
+// OpenAI Embedding Provider
+// =====================================================
+
+// OpenAIProvider implements EmbeddingProvider using OpenAI's API
+type OpenAIProvider struct {
+	apiKey     string
+	model      string
+	dimension  int
+	httpClient *http.Client
+}
+
+// NewOpenAIProvider creates a new OpenAI embedding provider
+func NewOpenAIProvider(apiKey, model string, dimension int) *OpenAIProvider {
+	return &OpenAIProvider{
+		apiKey:    apiKey,
+		model:     model,
+		dimension: dimension,
+		httpClient: &http.Client{
+			Timeout: 60 * time.Second,
+		},
+	}
+}
+
+// openAIEmbeddingRequest represents the OpenAI API request
+type openAIEmbeddingRequest struct {
+	Model      string   `json:"model"`
+	Input      []string `json:"input"`
+	Dimensions int      `json:"dimensions,omitempty"`
+}
+
+// openAIEmbeddingResponse represents the OpenAI API response
+type openAIEmbeddingResponse struct {
+	Data []struct {
+		Embedding []float32 `json:"embedding"`
+		Index     int       `json:"index"`
+	} `json:"data"`
+	Usage struct {
+		PromptTokens int `json:"prompt_tokens"`
+		TotalTokens  int `json:"total_tokens"`
+	} `json:"usage"`
+	Error *struct {
+		Message string `json:"message"`
+		Type    string `json:"type"`
+	} `json:"error,omitempty"`
+}
+
+// Embed generates embedding for a single text
+func (p *OpenAIProvider) Embed(ctx context.Context, text string) ([]float32, error) {
+	embeddings, err := p.EmbedBatch(ctx, []string{text})
+	if err != nil {
+		return nil, err
+	}
+	if len(embeddings) == 0 {
+		return nil, errors.New("no embedding returned")
+	}
+	return embeddings[0], nil
+}
+
+// EmbedBatch generates embeddings for multiple texts
+func (p *OpenAIProvider) EmbedBatch(ctx context.Context, texts []string) ([][]float32, error) {
+	if len(texts) == 0 {
+		return nil, nil
+	}
+
+	// Truncate texts to avoid token limits (max ~8000 tokens per text)
+	truncatedTexts := make([]string, len(texts))
+	for i, text := range texts {
+		if len(text) > 30000 { // Rough estimate: ~4 chars per token
+			truncatedTexts[i] = text[:30000]
+		} else {
+			truncatedTexts[i] = text
+		}
+	}
+
+	reqBody := openAIEmbeddingRequest{
+		Model: p.model,
+		Input: truncatedTexts,
+	}
+
+	// Only set dimensions for models that support it (text-embedding-3-*)
+	if p.model == "text-embedding-3-small" || p.model == "text-embedding-3-large" {
+		reqBody.Dimensions = p.dimension
+	}
+
+	body, err := json.Marshal(reqBody)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal request: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", "https://api.openai.com/v1/embeddings", bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	req.Header.Set("Authorization", "Bearer "+p.apiKey)
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := p.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("failed to call OpenAI API: %w", err)
+	}
+	defer resp.Body.Close()
+
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read response: %w", err)
+	}
+
+	var apiResp openAIEmbeddingResponse
+	if err := json.Unmarshal(respBody, &apiResp); err != nil {
+		return nil, fmt.Errorf("failed to parse response: %w", err)
+	}
+
+	if apiResp.Error != nil {
+		return nil, fmt.Errorf("OpenAI API error: %s", apiResp.Error.Message)
+	}
+
+	if len(apiResp.Data) != len(texts) {
+		return nil, fmt.Errorf("expected %d embeddings, got %d", len(texts), len(apiResp.Data))
+	}
+
+	// Sort by index to maintain order
+	result := make([][]float32, len(texts))
+	for _, item := range apiResp.Data {
+		result[item.Index] = item.Embedding
+	}
+
+	return result, nil
+}
+
+// Dimension returns the embedding dimension
+func (p *OpenAIProvider) Dimension() int {
+	return p.dimension
+}
+
+// =====================================================
+// Ollama Embedding Provider (for local models)
+// =====================================================
+
+// OllamaProvider implements EmbeddingProvider using Ollama's API
+type OllamaProvider struct {
+	baseURL    string
+	model      string
+	dimension  int
+	httpClient *http.Client
+}
+
+// NewOllamaProvider creates a new Ollama embedding provider
+func NewOllamaProvider(baseURL, model string, dimension int) (*OllamaProvider, error) {
+	return &OllamaProvider{
+		baseURL:   baseURL,
+		model:     model,
+		dimension: dimension,
+		httpClient: &http.Client{
+			Timeout: 120 * time.Second, // Ollama can be slow on first inference
+		},
+	}, nil
+}
+
+// ollamaEmbeddingRequest represents the Ollama API request
+type ollamaEmbeddingRequest struct {
+	Model  string `json:"model"`
+	Prompt string `json:"prompt"`
+}
+
+// ollamaEmbeddingResponse represents the Ollama API response
+type ollamaEmbeddingResponse struct {
+	Embedding []float32 `json:"embedding"`
+}
+
+// Embed generates embedding for a single text
+func (p *OllamaProvider) Embed(ctx context.Context, text string) ([]float32, error) {
+	// Truncate text
+	if len(text) > 30000 {
+		text = text[:30000]
+	}
+
+	reqBody := ollamaEmbeddingRequest{
+		Model:  p.model,
+		Prompt: text,
+	}
+
+	body, err := json.Marshal(reqBody)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal request: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", p.baseURL+"/api/embeddings", bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := p.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("failed to call Ollama API: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		respBody, _ := io.ReadAll(resp.Body)
+		return nil, fmt.Errorf("Ollama API error (status %d): %s", resp.StatusCode, string(respBody))
+	}
+
+	var apiResp ollamaEmbeddingResponse
+	if err := json.NewDecoder(resp.Body).Decode(&apiResp); err != nil {
+		return nil, fmt.Errorf("failed to parse response: %w", err)
+	}
+
+	return apiResp.Embedding, nil
+}
+
+// EmbedBatch generates embeddings for multiple texts (sequential for Ollama)
+func (p *OllamaProvider) EmbedBatch(ctx context.Context, texts []string) ([][]float32, error) {
+	result := make([][]float32, len(texts))
+
+	for i, text := range texts {
+		embedding, err := p.Embed(ctx, text)
+		if err != nil {
+			return nil, fmt.Errorf("failed to embed text %d: %w", i, err)
+		}
+		result[i] = embedding
+	}
+
+	return result, nil
+}
+
+// Dimension returns the embedding dimension
+func (p *OllamaProvider) Dimension() int {
+	return p.dimension
+}
--- a/edu-search-service/internal/embedding/embedding_test.go
+++ b/edu-search-service/internal/embedding/embedding_test.go
@@ -0,0 +1,319 @@
+package embedding
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+)
+
+func TestNewService_Disabled(t *testing.T) {
+	service, err := NewService("none", "", "", "", 1536, false)
+	if err != nil {
+		t.Fatalf("NewService failed: %v", err)
+	}
+
+	if service.IsEnabled() {
+		t.Error("Service should not be enabled")
+	}
+
+	if service.Dimension() != 1536 {
+		t.Errorf("Expected dimension 1536, got %d", service.Dimension())
+	}
+}
+
+func TestNewService_DisabledByProvider(t *testing.T) {
+	service, err := NewService("none", "", "", "", 1536, true)
+	if err != nil {
+		t.Fatalf("NewService failed: %v", err)
+	}
+
+	if service.IsEnabled() {
+		t.Error("Service should not be enabled when provider is 'none'")
+	}
+}
+
+func TestNewService_OpenAIMissingKey(t *testing.T) {
+	_, err := NewService("openai", "", "", "", 1536, true)
+	if err == nil {
+		t.Error("Expected error for missing OpenAI API key")
+	}
+}
+
+func TestNewService_UnknownProvider(t *testing.T) {
+	_, err := NewService("unknown", "", "", "", 1536, true)
+	if err == nil {
+		t.Error("Expected error for unknown provider")
+	}
+}
+
+func TestService_EmbedWhenDisabled(t *testing.T) {
+	service, _ := NewService("none", "", "", "", 1536, false)
+
+	_, err := service.Embed(context.Background(), "test text")
+	if err == nil {
+		t.Error("Expected error when embedding with disabled service")
+	}
+}
+
+func TestService_EmbedBatchWhenDisabled(t *testing.T) {
+	service, _ := NewService("none", "", "", "", 1536, false)
+
+	_, err := service.EmbedBatch(context.Background(), []string{"test1", "test2"})
+	if err == nil {
+		t.Error("Expected error when embedding batch with disabled service")
+	}
+}
+
+// =====================================================
+// OpenAI Provider Tests with Mock Server
+// =====================================================
+
+func TestOpenAIProvider_Embed(t *testing.T) {
+	// Create mock server
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// Verify request
+		if r.Method != "POST" {
+			t.Errorf("Expected POST, got %s", r.Method)
+		}
+		if r.Header.Get("Authorization") != "Bearer test-api-key" {
+			t.Errorf("Expected correct Authorization header")
+		}
+		if r.Header.Get("Content-Type") != "application/json" {
+			t.Errorf("Expected Content-Type application/json")
+		}
+
+		// Parse request body
+		var reqBody openAIEmbeddingRequest
+		if err := json.NewDecoder(r.Body).Decode(&reqBody); err != nil {
+			t.Fatalf("Failed to parse request body: %v", err)
+		}
+
+		if reqBody.Model != "text-embedding-3-small" {
+			t.Errorf("Expected model text-embedding-3-small, got %s", reqBody.Model)
+		}
+
+		// Send mock response
+		resp := openAIEmbeddingResponse{
+			Data: []struct {
+				Embedding []float32 `json:"embedding"`
+				Index     int       `json:"index"`
+			}{
+				{
+					Embedding: make([]float32, 1536),
+					Index:     0,
+				},
+			},
+		}
+		resp.Data[0].Embedding[0] = 0.1
+		resp.Data[0].Embedding[1] = 0.2
+
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(resp)
+	}))
+	defer server.Close()
+
+	// Create provider with mock server (we need to override the URL)
+	provider := &OpenAIProvider{
+		apiKey:    "test-api-key",
+		model:     "text-embedding-3-small",
+		dimension: 1536,
+		httpClient: &http.Client{
+			Timeout: 10 * time.Second,
+		},
+	}
+
+	// Note: This test won't actually work with the mock server because
+	// the provider hardcodes the OpenAI URL. This is a structural test.
+	// For real testing, we'd need to make the URL configurable.
+
+	if provider.Dimension() != 1536 {
+		t.Errorf("Expected dimension 1536, got %d", provider.Dimension())
+	}
+}
+
+func TestOpenAIProvider_EmbedBatch_EmptyInput(t *testing.T) {
+	provider := NewOpenAIProvider("test-key", "text-embedding-3-small", 1536)
+
+	result, err := provider.EmbedBatch(context.Background(), []string{})
+	if err != nil {
+		t.Errorf("Empty input should not cause error: %v", err)
+	}
+	if result != nil {
+		t.Errorf("Expected nil result for empty input, got %v", result)
+	}
+}
+
+// =====================================================
+// Ollama Provider Tests with Mock Server
+// =====================================================
+
+func TestOllamaProvider_Embed(t *testing.T) {
+	// Create mock server
+	mockEmbedding := make([]float32, 384)
+	mockEmbedding[0] = 0.5
+	mockEmbedding[1] = 0.3
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != "POST" {
+			t.Errorf("Expected POST, got %s", r.Method)
+		}
+		if r.URL.Path != "/api/embeddings" {
+			t.Errorf("Expected path /api/embeddings, got %s", r.URL.Path)
+		}
+
+		// Parse request
+		var reqBody ollamaEmbeddingRequest
+		if err := json.NewDecoder(r.Body).Decode(&reqBody); err != nil {
+			t.Fatalf("Failed to parse request: %v", err)
+		}
+
+		if reqBody.Model != "nomic-embed-text" {
+			t.Errorf("Expected model nomic-embed-text, got %s", reqBody.Model)
+		}
+
+		// Send response
+		resp := ollamaEmbeddingResponse{
+			Embedding: mockEmbedding,
+		}
+
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(resp)
+	}))
+	defer server.Close()
+
+	provider, err := NewOllamaProvider(server.URL, "nomic-embed-text", 384)
+	if err != nil {
+		t.Fatalf("Failed to create provider: %v", err)
+	}
+
+	ctx := context.Background()
+	embedding, err := provider.Embed(ctx, "Test text für Embedding")
+
+	if err != nil {
+		t.Fatalf("Embed failed: %v", err)
+	}
+
+	if len(embedding) != 384 {
+		t.Errorf("Expected 384 dimensions, got %d", len(embedding))
+	}
+
+	if embedding[0] != 0.5 {
+		t.Errorf("Expected first value 0.5, got %f", embedding[0])
+	}
+}
+
+func TestOllamaProvider_EmbedBatch(t *testing.T) {
+	callCount := 0
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		callCount++
+
+		mockEmbedding := make([]float32, 384)
+		mockEmbedding[0] = float32(callCount) * 0.1
+
+		resp := ollamaEmbeddingResponse{
+			Embedding: mockEmbedding,
+		}
+
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(resp)
+	}))
+	defer server.Close()
+
+	provider, err := NewOllamaProvider(server.URL, "nomic-embed-text", 384)
+	if err != nil {
+		t.Fatalf("Failed to create provider: %v", err)
+	}
+
+	ctx := context.Background()
+	texts := []string{"Text 1", "Text 2", "Text 3"}
+	embeddings, err := provider.EmbedBatch(ctx, texts)
+
+	if err != nil {
+		t.Fatalf("EmbedBatch failed: %v", err)
+	}
+
+	if len(embeddings) != 3 {
+		t.Errorf("Expected 3 embeddings, got %d", len(embeddings))
+	}
+
+	// Verify each embedding was called
+	if callCount != 3 {
+		t.Errorf("Expected 3 API calls, got %d", callCount)
+	}
+}
+
+func TestOllamaProvider_EmbedServerError(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+		w.Write([]byte("Internal server error"))
+	}))
+	defer server.Close()
+
+	provider, _ := NewOllamaProvider(server.URL, "nomic-embed-text", 384)
+
+	_, err := provider.Embed(context.Background(), "test")
+	if err == nil {
+		t.Error("Expected error for server error response")
+	}
+}
+
+func TestOllamaProvider_Dimension(t *testing.T) {
+	provider, _ := NewOllamaProvider("http://localhost:11434", "nomic-embed-text", 768)
+
+	if provider.Dimension() != 768 {
+		t.Errorf("Expected dimension 768, got %d", provider.Dimension())
+	}
+}
+
+// =====================================================
+// Text Truncation Tests
+// =====================================================
+
+func TestOllamaProvider_TextTruncation(t *testing.T) {
+	receivedText := ""
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var reqBody ollamaEmbeddingRequest
+		json.NewDecoder(r.Body).Decode(&reqBody)
+		receivedText = reqBody.Prompt
+
+		resp := ollamaEmbeddingResponse{
+			Embedding: make([]float32, 384),
+		}
+		w.Header().Set("Content-Type", "application/json")
+		json.NewEncoder(w).Encode(resp)
+	}))
+	defer server.Close()
+
+	provider, _ := NewOllamaProvider(server.URL, "nomic-embed-text", 384)
+
+	// Create very long text
+	longText := ""
+	for i := 0; i < 40000; i++ {
+		longText += "a"
+	}
+
+	provider.Embed(context.Background(), longText)
+
+	// Text should be truncated to 30000 chars
+	if len(receivedText) > 30000 {
+		t.Errorf("Expected truncated text <= 30000 chars, got %d", len(receivedText))
+	}
+}
+
+// =====================================================
+// Integration Tests (require actual service)
+// =====================================================
+
+func TestOpenAIProvider_Integration(t *testing.T) {
+	// Skip in CI/CD - only run manually with real API key
+	t.Skip("Integration test - requires OPENAI_API_KEY environment variable")
+
+	// provider := NewOpenAIProvider(os.Getenv("OPENAI_API_KEY"), "text-embedding-3-small", 1536)
+	// embedding, err := provider.Embed(context.Background(), "Lehrplan Mathematik Bayern")
+	// ...
+}
--- a/edu-search-service/internal/extractor/extractor.go
+++ b/edu-search-service/internal/extractor/extractor.go
@@ -0,0 +1,464 @@
+package extractor
+
+import (
+	"bytes"
+	"io"
+	"regexp"
+	"strings"
+	"unicode"
+
+	"github.com/PuerkitoBio/goquery"
+	"github.com/ledongthuc/pdf"
+	"golang.org/x/net/html"
+)
+
+// ExtractedContent contains parsed content from HTML/PDF
+type ExtractedContent struct {
+	Title         string
+	ContentText   string
+	SnippetText   string
+	Language      string
+	ContentLength int
+	Headings      []string
+	Links         []string
+	MetaData      map[string]string
+	Features      ContentFeatures
+}
+
+// ContentFeatures for quality scoring
+type ContentFeatures struct {
+	AdDensity       float64
+	LinkDensity     float64
+	TextToHTMLRatio float64
+	HasMainContent  bool
+}
+
+// ExtractHTML extracts content from HTML
+func ExtractHTML(body []byte) (*ExtractedContent, error) {
+	doc, err := goquery.NewDocumentFromReader(bytes.NewReader(body))
+	if err != nil {
+		return nil, err
+	}
+
+	content := &ExtractedContent{
+		MetaData: make(map[string]string),
+	}
+
+	// Extract title
+	content.Title = strings.TrimSpace(doc.Find("title").First().Text())
+	if content.Title == "" {
+		content.Title = strings.TrimSpace(doc.Find("h1").First().Text())
+	}
+
+	// Extract meta tags
+	doc.Find("meta").Each(func(i int, s *goquery.Selection) {
+		name, _ := s.Attr("name")
+		property, _ := s.Attr("property")
+		contentAttr, _ := s.Attr("content")
+
+		key := name
+		if key == "" {
+			key = property
+		}
+
+		if key != "" && contentAttr != "" {
+			content.MetaData[strings.ToLower(key)] = contentAttr
+		}
+	})
+
+	// Try to get og:title if main title is empty
+	if content.Title == "" {
+		if ogTitle, ok := content.MetaData["og:title"]; ok {
+			content.Title = ogTitle
+		}
+	}
+
+	// Extract headings
+	doc.Find("h1, h2, h3").Each(func(i int, s *goquery.Selection) {
+		text := strings.TrimSpace(s.Text())
+		if text != "" && len(text) < 500 {
+			content.Headings = append(content.Headings, text)
+		}
+	})
+
+	// Remove unwanted elements
+	doc.Find("script, style, nav, header, footer, aside, iframe, noscript, form, .advertisement, .ad, .ads, #cookie-banner, .cookie-notice, .social-share").Remove()
+
+	// Try to find main content area
+	mainContent := doc.Find("main, article, .content, .main-content, #content, #main").First()
+	if mainContent.Length() == 0 {
+		mainContent = doc.Find("body")
+	}
+
+	// Extract text content
+	var textBuilder strings.Builder
+	mainContent.Find("p, li, td, th, h1, h2, h3, h4, h5, h6, blockquote, pre").Each(func(i int, s *goquery.Selection) {
+		text := strings.TrimSpace(s.Text())
+		if text != "" {
+			textBuilder.WriteString(text)
+			textBuilder.WriteString("\n\n")
+		}
+	})
+
+	content.ContentText = cleanText(textBuilder.String())
+	content.ContentLength = len(content.ContentText)
+
+	// Generate snippet (first ~300 chars of meaningful content)
+	content.SnippetText = generateSnippet(content.ContentText, 300)
+
+	// Extract links
+	doc.Find("a[href]").Each(func(i int, s *goquery.Selection) {
+		href, exists := s.Attr("href")
+		if exists && strings.HasPrefix(href, "http") {
+			content.Links = append(content.Links, href)
+		}
+	})
+
+	// Detect language
+	content.Language = detectLanguage(content.ContentText, content.MetaData)
+
+	// Calculate features
+	htmlLen := float64(len(body))
+	textLen := float64(len(content.ContentText))
+
+	if htmlLen > 0 {
+		content.Features.TextToHTMLRatio = textLen / htmlLen
+	}
+
+	if textLen > 0 {
+		linkTextLen := 0.0
+		doc.Find("a").Each(func(i int, s *goquery.Selection) {
+			linkTextLen += float64(len(s.Text()))
+		})
+		content.Features.LinkDensity = linkTextLen / textLen
+	}
+
+	content.Features.HasMainContent = content.ContentLength > 200
+
+	// Ad density estimation (very simple heuristic)
+	adCount := doc.Find(".ad, .ads, .advertisement, [class*='banner'], [id*='banner']").Length()
+	totalElements := doc.Find("div, p, article, section").Length()
+	if totalElements > 0 {
+		content.Features.AdDensity = float64(adCount) / float64(totalElements)
+	}
+
+	return content, nil
+}
+
+// ExtractPDF extracts text from PDF using ledongthuc/pdf library
+func ExtractPDF(body []byte) (*ExtractedContent, error) {
+	content := &ExtractedContent{
+		MetaData: make(map[string]string),
+	}
+
+	// Create a reader from the byte slice
+	reader := bytes.NewReader(body)
+	pdfReader, err := pdf.NewReader(reader, int64(len(body)))
+	if err != nil {
+		// Fallback to basic extraction if PDF parsing fails
+		return extractPDFFallback(body)
+	}
+
+	// Extract text using GetPlainText
+	textReader, err := pdfReader.GetPlainText()
+	if err != nil {
+		// Fallback to basic extraction
+		return extractPDFFallback(body)
+	}
+
+	// Read all text content
+	var textBuilder strings.Builder
+	_, err = io.Copy(&textBuilder, textReader)
+	if err != nil {
+		return extractPDFFallback(body)
+	}
+
+	rawText := textBuilder.String()
+
+	// Clean and process text
+	content.ContentText = cleanText(rawText)
+	content.ContentLength = len(content.ContentText)
+	content.SnippetText = generateSnippet(content.ContentText, 300)
+	content.Language = detectLanguage(content.ContentText, nil)
+	content.Features.HasMainContent = content.ContentLength > 200
+
+	// Extract title from first significant line
+	content.Title = extractPDFTitle(content.ContentText)
+
+	// Try to extract headings (larger font text often appears first in lines)
+	content.Headings = extractPDFHeadings(content.ContentText)
+
+	// Set PDF-specific metadata
+	content.MetaData["content_type"] = "application/pdf"
+	content.MetaData["page_count"] = string(rune(pdfReader.NumPage()))
+
+	return content, nil
+}
+
+// ExtractPDFWithMetadata extracts text with page-by-page processing
+// Use this when you need more control over the extraction process
+func ExtractPDFWithMetadata(body []byte) (*ExtractedContent, error) {
+	content := &ExtractedContent{
+		MetaData: make(map[string]string),
+	}
+
+	reader := bytes.NewReader(body)
+	pdfReader, err := pdf.NewReader(reader, int64(len(body)))
+	if err != nil {
+		return extractPDFFallback(body)
+	}
+
+	// Extract text page by page for better control
+	var textBuilder strings.Builder
+	numPages := pdfReader.NumPage()
+
+	for pageNum := 1; pageNum <= numPages; pageNum++ {
+		page := pdfReader.Page(pageNum)
+		if page.V.IsNull() {
+			continue
+		}
+
+		// Get page content
+		pageContent := page.Content()
+		for _, text := range pageContent.Text {
+			textBuilder.WriteString(text.S)
+			textBuilder.WriteString(" ")
+		}
+		textBuilder.WriteString("\n")
+	}
+
+	rawText := textBuilder.String()
+
+	// Clean and process text
+	content.ContentText = cleanText(rawText)
+	content.ContentLength = len(content.ContentText)
+	content.SnippetText = generateSnippet(content.ContentText, 300)
+	content.Language = detectLanguage(content.ContentText, nil)
+	content.Features.HasMainContent = content.ContentLength > 200
+
+	// Extract title and headings from plain text
+	content.Title = extractPDFTitle(content.ContentText)
+	content.Headings = extractPDFHeadings(content.ContentText)
+
+	content.MetaData["content_type"] = "application/pdf"
+	content.MetaData["page_count"] = string(rune(numPages))
+	content.MetaData["extraction_method"] = "page_by_page"
+
+	return content, nil
+}
+
+// extractPDFFallback uses basic regex extraction when PDF library fails
+func extractPDFFallback(body []byte) (*ExtractedContent, error) {
+	content := &ExtractedContent{
+		MetaData: make(map[string]string),
+	}
+
+	// Basic PDF text extraction using regex (fallback)
+	pdfContent := string(body)
+	var textBuilder strings.Builder
+
+	// Find text content in PDF streams
+	re := regexp.MustCompile(`\((.*?)\)`)
+	matches := re.FindAllStringSubmatch(pdfContent, -1)
+
+	for _, match := range matches {
+		if len(match) > 1 {
+			text := match[1]
+			if isPrintableText(text) {
+				textBuilder.WriteString(text)
+				textBuilder.WriteString(" ")
+			}
+		}
+	}
+
+	content.ContentText = cleanText(textBuilder.String())
+	content.ContentLength = len(content.ContentText)
+	content.SnippetText = generateSnippet(content.ContentText, 300)
+	content.Language = detectLanguage(content.ContentText, nil)
+	content.Features.HasMainContent = content.ContentLength > 200
+	content.Title = extractPDFTitle(content.ContentText)
+	content.MetaData["content_type"] = "application/pdf"
+	content.MetaData["extraction_method"] = "fallback"
+
+	return content, nil
+}
+
+// extractPDFTitle extracts title from PDF content (first significant line)
+func extractPDFTitle(text string) string {
+	lines := strings.Split(text, "\n")
+	for _, line := range lines {
+		line = strings.TrimSpace(line)
+		// Title should be meaningful length
+		if len(line) >= 10 && len(line) <= 200 {
+			// Skip lines that look like page numbers or dates
+			if !regexp.MustCompile(`^\d+$`).MatchString(line) &&
+				!regexp.MustCompile(`^\d{1,2}\.\d{1,2}\.\d{2,4}$`).MatchString(line) {
+				return line
+			}
+		}
+	}
+	return ""
+}
+
+// extractPDFHeadings attempts to extract headings from plain text
+func extractPDFHeadings(text string) []string {
+	var headings []string
+	lines := strings.Split(text, "\n")
+
+	for i, line := range lines {
+		line = strings.TrimSpace(line)
+		// Skip very short or very long lines
+		if len(line) < 5 || len(line) > 200 {
+			continue
+		}
+
+		// Heuristics for headings:
+		// 1. All caps lines (common in PDFs)
+		// 2. Lines followed by empty line or starting with numbers (1., 1.1, etc.)
+		// 3. Short lines at beginning of document
+
+		isAllCaps := line == strings.ToUpper(line) && strings.ContainsAny(line, "ABCDEFGHIJKLMNOPQRSTUVWXYZÄÖÜ")
+		isNumbered := regexp.MustCompile(`^\d+(\.\d+)*\.?\s+\S`).MatchString(line)
+		isShortAndEarly := i < 20 && len(line) < 80
+
+		if (isAllCaps || isNumbered || isShortAndEarly) && !containsHeading(headings, line) {
+			headings = append(headings, line)
+			if len(headings) >= 10 {
+				break // Limit to 10 headings
+			}
+		}
+	}
+
+	return headings
+}
+
+// containsHeading checks if a heading already exists in the list
+func containsHeading(headings []string, heading string) bool {
+	for _, h := range headings {
+		if h == heading {
+			return true
+		}
+	}
+	return false
+}
+
+func isPrintableText(s string) bool {
+	if len(s) < 3 {
+		return false
+	}
+
+	printable := 0
+	for _, r := range s {
+		if unicode.IsPrint(r) && (unicode.IsLetter(r) || unicode.IsSpace(r) || unicode.IsPunct(r)) {
+			printable++
+		}
+	}
+
+	return float64(printable)/float64(len(s)) > 0.7
+}
+
+func cleanText(text string) string {
+	// Normalize whitespace
+	text = strings.ReplaceAll(text, "\r\n", "\n")
+	text = strings.ReplaceAll(text, "\r", "\n")
+
+	// Replace multiple newlines with double newline
+	re := regexp.MustCompile(`\n{3,}`)
+	text = re.ReplaceAllString(text, "\n\n")
+
+	// Replace multiple spaces with single space
+	re = regexp.MustCompile(`[ \t]+`)
+	text = re.ReplaceAllString(text, " ")
+
+	// Trim each line
+	lines := strings.Split(text, "\n")
+	for i, line := range lines {
+		lines[i] = strings.TrimSpace(line)
+	}
+	text = strings.Join(lines, "\n")
+
+	return strings.TrimSpace(text)
+}
+
+func generateSnippet(text string, maxLen int) string {
+	// Find first paragraph with enough content
+	paragraphs := strings.Split(text, "\n\n")
+
+	for _, p := range paragraphs {
+		p = strings.TrimSpace(p)
+		if len(p) >= 50 {
+			if len(p) > maxLen {
+				// Find word boundary
+				p = p[:maxLen]
+				lastSpace := strings.LastIndex(p, " ")
+				if lastSpace > maxLen/2 {
+					p = p[:lastSpace]
+				}
+				p += "..."
+			}
+			return p
+		}
+	}
+
+	// Fallback: just truncate
+	if len(text) > maxLen {
+		text = text[:maxLen] + "..."
+	}
+	return text
+}
+
+func detectLanguage(text string, meta map[string]string) string {
+	// Check meta tags first
+	if meta != nil {
+		if lang, ok := meta["og:locale"]; ok {
+			if strings.HasPrefix(lang, "de") {
+				return "de"
+			}
+			if strings.HasPrefix(lang, "en") {
+				return "en"
+			}
+		}
+	}
+
+	// Simple heuristic based on common German words
+	germanWords := []string{
+		"und", "der", "die", "das", "ist", "für", "mit", "von",
+		"werden", "wird", "sind", "auch", "als", "können", "nach",
+		"einer", "durch", "sich", "bei", "sein", "noch", "haben",
+	}
+
+	englishWords := []string{
+		"the", "and", "for", "are", "but", "not", "you", "all",
+		"can", "had", "her", "was", "one", "our", "with", "they",
+	}
+
+	lowerText := strings.ToLower(text)
+
+	germanCount := 0
+	for _, word := range germanWords {
+		if strings.Contains(lowerText, " "+word+" ") {
+			germanCount++
+		}
+	}
+
+	englishCount := 0
+	for _, word := range englishWords {
+		if strings.Contains(lowerText, " "+word+" ") {
+			englishCount++
+		}
+	}
+
+	if germanCount > englishCount && germanCount > 3 {
+		return "de"
+	}
+	if englishCount > germanCount && englishCount > 3 {
+		return "en"
+	}
+
+	return "de" // Default to German for education content
+}
+
+// UnescapeHTML unescapes HTML entities
+func UnescapeHTML(s string) string {
+	return html.UnescapeString(s)
+}
--- a/edu-search-service/internal/extractor/extractor_test.go
+++ b/edu-search-service/internal/extractor/extractor_test.go
@@ -0,0 +1,802 @@
+package extractor
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestExtractHTML_BasicContent(t *testing.T) {
+	html := []byte(`<!DOCTYPE html>
+<html>
+<head>
+	<title>Test Page Title</title>
+	<meta name="description" content="Test description">
+	<meta property="og:title" content="OG Title">
+</head>
+<body>
+	<h1>Main Heading</h1>
+	<p>This is the first paragraph with some meaningful content.</p>
+	<p>This is another paragraph that adds more information.</p>
+</body>
+</html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatalf("ExtractHTML failed: %v", err)
+	}
+
+	// Check title
+	if content.Title != "Test Page Title" {
+		t.Errorf("Expected title 'Test Page Title', got %q", content.Title)
+	}
+
+	// Check metadata
+	if content.MetaData["description"] != "Test description" {
+		t.Errorf("Expected description 'Test description', got %q", content.MetaData["description"])
+	}
+
+	// Check headings
+	if len(content.Headings) == 0 {
+		t.Error("Expected at least one heading")
+	}
+	if content.Headings[0] != "Main Heading" {
+		t.Errorf("Expected heading 'Main Heading', got %q", content.Headings[0])
+	}
+
+	// Check content text
+	if !strings.Contains(content.ContentText, "first paragraph") {
+		t.Error("Expected content to contain 'first paragraph'")
+	}
+}
+
+func TestExtractHTML_TitleFallback(t *testing.T) {
+	tests := []struct {
+		name     string
+		html     string
+		expected string
+	}{
+		{
+			name: "Title from title tag",
+			html: `<html><head><title>Page Title</title></head><body></body></html>`,
+			expected: "Page Title",
+		},
+		{
+			name: "Title from H1 when no title tag",
+			html: `<html><head></head><body><h1>H1 Title</h1></body></html>`,
+			expected: "H1 Title",
+		},
+		{
+			name: "Title from og:title when no title or h1",
+			html: `<html><head><meta property="og:title" content="OG Title"></head><body></body></html>`,
+			expected: "OG Title",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			content, err := ExtractHTML([]byte(tt.html))
+			if err != nil {
+				t.Fatalf("ExtractHTML failed: %v", err)
+			}
+			if content.Title != tt.expected {
+				t.Errorf("Expected title %q, got %q", tt.expected, content.Title)
+			}
+		})
+	}
+}
+
+func TestExtractHTML_RemovesUnwantedElements(t *testing.T) {
+	html := []byte(`<html>
+<body>
+	<nav>Navigation menu</nav>
+	<header>Header content</header>
+	<main>
+		<p>Main content paragraph</p>
+	</main>
+	<script>alert('dangerous');</script>
+	<style>.hidden{display:none;}</style>
+	<footer>Footer content</footer>
+	<aside>Sidebar content</aside>
+	<div class="advertisement">Ad content</div>
+</body>
+</html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Should contain main content
+	if !strings.Contains(content.ContentText, "Main content paragraph") {
+		t.Error("Expected main content to be extracted")
+	}
+
+	// Should not contain unwanted elements
+	unwanted := []string{"Navigation menu", "alert('dangerous')", "Footer content", "Ad content"}
+	for _, text := range unwanted {
+		if strings.Contains(content.ContentText, text) {
+			t.Errorf("Content should not contain %q", text)
+		}
+	}
+}
+
+func TestExtractHTML_ExtractsLinks(t *testing.T) {
+	html := []byte(`<html><body>
+	<a href="https://example.com/page1">Link 1</a>
+	<a href="https://example.com/page2">Link 2</a>
+	<a href="/relative/path">Relative Link</a>
+	<a href="mailto:test@example.com">Email</a>
+</body></html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Should extract absolute HTTP links
+	if len(content.Links) != 2 {
+		t.Errorf("Expected 2 HTTP links, got %d", len(content.Links))
+	}
+
+	hasPage1 := false
+	hasPage2 := false
+	for _, link := range content.Links {
+		if link == "https://example.com/page1" {
+			hasPage1 = true
+		}
+		if link == "https://example.com/page2" {
+			hasPage2 = true
+		}
+	}
+
+	if !hasPage1 || !hasPage2 {
+		t.Error("Expected to find both HTTP links")
+	}
+}
+
+func TestExtractHTML_CalculatesFeatures(t *testing.T) {
+	html := []byte(`<html><body>
+	<div class="advertisement">Ad 1</div>
+	<p>Some content text that is long enough to be meaningful and provide a good ratio.</p>
+	<p>More content here to increase the text length.</p>
+	<a href="#">Link 1</a>
+	<a href="#">Link 2</a>
+</body></html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Check features are calculated
+	if content.Features.TextToHTMLRatio <= 0 {
+		t.Error("Expected positive TextToHTMLRatio")
+	}
+
+	// Content should have length
+	if content.ContentLength == 0 {
+		t.Error("Expected non-zero ContentLength")
+	}
+}
+
+func TestExtractHTML_GeneratesSnippet(t *testing.T) {
+	html := []byte(`<html><body>
+	<p>This is a short intro.</p>
+	<p>This is a longer paragraph that should be used as the snippet because it has more meaningful content and meets the minimum length requirement for a good snippet.</p>
+	<p>Another paragraph here.</p>
+</body></html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if content.SnippetText == "" {
+		t.Error("Expected non-empty snippet")
+	}
+
+	// Snippet should be limited in length
+	if len(content.SnippetText) > 350 { // 300 + "..." margin
+		t.Errorf("Snippet too long: %d chars", len(content.SnippetText))
+	}
+}
+
+func TestDetectLanguage(t *testing.T) {
+	tests := []struct {
+		name     string
+		text     string
+		meta     map[string]string
+		expected string
+	}{
+		{
+			name:     "German from meta",
+			text:     "Some text",
+			meta:     map[string]string{"og:locale": "de_DE"},
+			expected: "de",
+		},
+		{
+			name:     "English from meta",
+			text:     "Some text",
+			meta:     map[string]string{"og:locale": "en_US"},
+			expected: "en",
+		},
+		{
+			name:     "German from content",
+			text:     "Dies ist ein Text und der Inhalt wird hier analysiert",
+			meta:     nil,
+			expected: "de",
+		},
+		{
+			name:     "English from content",
+			text:     "This is the content and we are analyzing the text here with all the words they can use for things but not any German",
+			meta:     nil,
+			expected: "en",
+		},
+		{
+			name:     "Default to German for ambiguous",
+			text:     "Hello World",
+			meta:     nil,
+			expected: "de",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := detectLanguage(tt.text, tt.meta)
+			if result != tt.expected {
+				t.Errorf("detectLanguage() = %q, expected %q", result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestCleanText(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name:     "Normalize Windows line endings",
+			input:    "Line1\r\nLine2",
+			expected: "Line1\nLine2",
+		},
+		{
+			name:     "Collapse multiple newlines",
+			input:    "Line1\n\n\n\n\nLine2",
+			expected: "Line1\n\nLine2",
+		},
+		{
+			name:     "Collapse multiple spaces",
+			input:    "Word1     Word2",
+			expected: "Word1 Word2",
+		},
+		{
+			name:     "Trim whitespace",
+			input:    "  Text with spaces  \n  More text  ",
+			expected: "Text with spaces\nMore text",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := cleanText(tt.input)
+			if result != tt.expected {
+				t.Errorf("cleanText(%q) = %q, expected %q", tt.input, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestGenerateSnippet(t *testing.T) {
+	tests := []struct {
+		name     string
+		text     string
+		maxLen   int
+		checkFn  func(string) bool
+	}{
+		{
+			name:   "Short text unchanged",
+			text:   "Short paragraph.",
+			maxLen: 300,
+			checkFn: func(s string) bool {
+				return s == "Short paragraph."
+			},
+		},
+		{
+			name:   "Long text truncated",
+			text:   strings.Repeat("A long sentence that keeps going. ", 20),
+			maxLen: 100,
+			checkFn: func(s string) bool {
+				return len(s) <= 103 && strings.HasSuffix(s, "...")
+			},
+		},
+		{
+			name: "First suitable paragraph",
+			text: "Tiny.\n\nThis is a paragraph with enough content to be used as a snippet because it meets the minimum length.",
+			maxLen: 300,
+			checkFn: func(s string) bool {
+				return strings.HasPrefix(s, "This is a paragraph")
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := generateSnippet(tt.text, tt.maxLen)
+			if !tt.checkFn(result) {
+				t.Errorf("generateSnippet() = %q, check failed", result)
+			}
+		})
+	}
+}
+
+func TestIsPrintableText(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected bool
+	}{
+		{
+			name:     "Normal text",
+			input:    "Hello World",
+			expected: true,
+		},
+		{
+			name:     "German text",
+			input:    "Übung mit Umlauten",
+			expected: true,
+		},
+		{
+			name:     "Too short",
+			input:    "AB",
+			expected: false,
+		},
+		{
+			name:     "Binary data",
+			input:    "\x00\x01\x02\x03\x04",
+			expected: false,
+		},
+		{
+			name:     "Mixed printable",
+			input:    "Text with some \x00 binary",
+			expected: true, // >70% printable
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := isPrintableText(tt.input)
+			if result != tt.expected {
+				t.Errorf("isPrintableText(%q) = %v, expected %v", tt.input, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestExtractHTML_HeadingsExtraction(t *testing.T) {
+	html := []byte(`<html><body>
+	<h1>Main Title</h1>
+	<h2>Section 1</h2>
+	<p>Content</p>
+	<h2>Section 2</h2>
+	<h3>Subsection 2.1</h3>
+	<p>More content</p>
+</body></html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if len(content.Headings) != 4 {
+		t.Errorf("Expected 4 headings (h1, h2, h2, h3), got %d", len(content.Headings))
+	}
+
+	expectedHeadings := []string{"Main Title", "Section 1", "Section 2", "Subsection 2.1"}
+	for i, expected := range expectedHeadings {
+		if i < len(content.Headings) && content.Headings[i] != expected {
+			t.Errorf("Heading %d: expected %q, got %q", i, expected, content.Headings[i])
+		}
+	}
+}
+
+func TestExtractHTML_ContentFromMain(t *testing.T) {
+	html := []byte(`<html><body>
+	<div>Outside main</div>
+	<main>
+		<article>
+			<p>Article content that is inside the main element.</p>
+		</article>
+	</main>
+	<div>Also outside</div>
+</body></html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if !strings.Contains(content.ContentText, "Article content") {
+		t.Error("Expected content from main element")
+	}
+}
+
+func TestExtractHTML_MetadataExtraction(t *testing.T) {
+	html := []byte(`<html>
+<head>
+	<meta name="author" content="Test Author">
+	<meta name="keywords" content="education, learning">
+	<meta property="og:description" content="OG Description">
+</head>
+<body></body>
+</html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if content.MetaData["author"] != "Test Author" {
+		t.Errorf("Expected author 'Test Author', got %q", content.MetaData["author"])
+	}
+	if content.MetaData["keywords"] != "education, learning" {
+		t.Errorf("Expected keywords, got %q", content.MetaData["keywords"])
+	}
+	if content.MetaData["og:description"] != "OG Description" {
+		t.Errorf("Expected og:description, got %q", content.MetaData["og:description"])
+	}
+}
+
+func TestUnescapeHTML(t *testing.T) {
+	tests := []struct {
+		input    string
+		expected string
+	}{
+		{"&amp;", "&"},
+		{"&lt;script&gt;", "<script>"},
+		{"&quot;quoted&quot;", "\"quoted\""},
+		{"&#39;apostrophe&#39;", "'apostrophe'"},
+		{"No entities", "No entities"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.input, func(t *testing.T) {
+			result := UnescapeHTML(tt.input)
+			if result != tt.expected {
+				t.Errorf("UnescapeHTML(%q) = %q, expected %q", tt.input, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestExtractPDF_BasicText(t *testing.T) {
+	// Create minimal PDF-like content with text markers
+	// Real PDFs would have proper structure, but we test the extraction logic
+	pdfContent := []byte("(Hello World) (This is a test)")
+
+	content, err := ExtractPDF(pdfContent)
+	if err != nil {
+		t.Fatalf("ExtractPDF failed: %v", err)
+	}
+
+	// Should extract some text
+	if content.ContentLength == 0 && !strings.Contains(string(pdfContent), "(Hello") {
+		// Only fail if there's actually extractable content
+		t.Log("PDF extraction returned empty content (expected for simple test)")
+	}
+
+	// Features should be set
+	if content.Language == "" {
+		t.Error("Expected language to be set")
+	}
+}
+
+func TestExtractHTML_AdDensity(t *testing.T) {
+	html := []byte(`<html><body>
+	<div class="advertisement">Ad 1</div>
+	<div class="advertisement">Ad 2</div>
+	<div class="advertisement">Ad 3</div>
+	<p>Content</p>
+	<div>Normal div</div>
+</body></html>`)
+
+	content, err := ExtractHTML(html)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Ad density should be calculated (3 ads / total divs)
+	if content.Features.AdDensity < 0 {
+		t.Error("AdDensity should not be negative")
+	}
+}
+
+func TestExtractHTML_HasMainContent(t *testing.T) {
+	tests := []struct {
+		name     string
+		html     string
+		expected bool
+	}{
+		{
+			name:     "Sufficient content",
+			html:     `<html><body><p>` + strings.Repeat("Content ", 50) + `</p></body></html>`,
+			expected: true,
+		},
+		{
+			name:     "Insufficient content",
+			html:     `<html><body><p>Short</p></body></html>`,
+			expected: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			content, err := ExtractHTML([]byte(tt.html))
+			if err != nil {
+				t.Fatal(err)
+			}
+			if content.Features.HasMainContent != tt.expected {
+				t.Errorf("HasMainContent = %v, expected %v", content.Features.HasMainContent, tt.expected)
+			}
+		})
+	}
+}
+
+// ============================================================
+// PDF Extraction Tests
+// ============================================================
+
+func TestExtractPDF_FallbackForInvalidPDF(t *testing.T) {
+	// Test with non-PDF content - should fallback gracefully
+	invalidPDF := []byte("This is not a PDF file (just some text content)")
+
+	content, err := ExtractPDF(invalidPDF)
+	if err != nil {
+		t.Fatalf("ExtractPDF should not fail completely: %v", err)
+	}
+
+	// Should still return a valid ExtractedContent struct
+	if content == nil {
+		t.Fatal("Expected non-nil content")
+	}
+
+	// Should detect fallback method
+	if content.MetaData["extraction_method"] != "fallback" {
+		t.Log("PDF fallback extraction was used as expected")
+	}
+}
+
+func TestExtractPDF_MetadataSet(t *testing.T) {
+	// Simple test content
+	content, err := ExtractPDF([]byte("(Test content)"))
+	if err != nil {
+		t.Fatalf("ExtractPDF failed: %v", err)
+	}
+
+	// Content type should be set
+	if content.MetaData["content_type"] != "application/pdf" {
+		t.Errorf("Expected content_type 'application/pdf', got %q", content.MetaData["content_type"])
+	}
+
+	// Language should be detected (default to German)
+	if content.Language == "" {
+		t.Error("Expected language to be set")
+	}
+}
+
+func TestExtractPDFTitle(t *testing.T) {
+	tests := []struct {
+		name     string
+		text     string
+		expected string
+	}{
+		{
+			name:     "Normal title",
+			text:     "Lehrplan Mathematik Bayern\n\nDieses Dokument beschreibt...",
+			expected: "Lehrplan Mathematik Bayern",
+		},
+		{
+			name:     "Skip page number",
+			text:     "1\n\nLehrplan Mathematik Bayern\n\nDieses Dokument...",
+			expected: "Lehrplan Mathematik Bayern",
+		},
+		{
+			name:     "Skip date",
+			text:     "15.01.2025\n\nLehrplan Mathematik\n\nDieses Dokument...",
+			expected: "Lehrplan Mathematik",
+		},
+		{
+			name:     "Skip short lines",
+			text:     "Short\n\nThis is a proper title for the document\n\nContent...",
+			expected: "This is a proper title for the document",
+		},
+		{
+			name:     "Empty text",
+			text:     "",
+			expected: "",
+		},
+		{
+			name:     "Only short lines",
+			text:     "A\nB\nC\nD",
+			expected: "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := extractPDFTitle(tt.text)
+			if result != tt.expected {
+				t.Errorf("extractPDFTitle() = %q, expected %q", result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestExtractPDFHeadings(t *testing.T) {
+	tests := []struct {
+		name            string
+		text            string
+		minHeadingCount int
+		expectedFirst   string
+	}{
+		{
+			name: "All caps headings",
+			text: `EINLEITUNG
+
+Dieser Text beschreibt die wichtigsten Punkte.
+
+KAPITEL EINS
+
+Hier folgt der erste Abschnitt.`,
+			minHeadingCount: 2,
+			expectedFirst:   "EINLEITUNG",
+		},
+		{
+			name: "Numbered headings",
+			text: `1. Einführung
+
+Text hier.
+
+1.1 Unterabschnitt
+
+Mehr Text.
+
+2. Hauptteil
+
+Weiterer Inhalt.`,
+			minHeadingCount: 3,
+			expectedFirst:   "1. Einführung",
+		},
+		{
+			name:            "No headings",
+			text:            "einfacher text ohne ueberschriften der nur aus kleinen buchstaben besteht und sehr lang ist damit er nicht als ueberschrift erkannt wird",
+			minHeadingCount: 0,
+			expectedFirst:   "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			headings := extractPDFHeadings(tt.text)
+
+			if len(headings) < tt.minHeadingCount {
+				t.Errorf("Expected at least %d headings, got %d", tt.minHeadingCount, len(headings))
+			}
+
+			if tt.expectedFirst != "" && len(headings) > 0 && headings[0] != tt.expectedFirst {
+				t.Errorf("Expected first heading %q, got %q", tt.expectedFirst, headings[0])
+			}
+		})
+	}
+}
+
+func TestExtractPDFHeadings_Limit(t *testing.T) {
+	// Test that headings are limited to 10
+	text := ""
+	for i := 1; i <= 20; i++ {
+		text += "KAPITEL " + strings.Repeat("X", i) + "\n\nText Text Text.\n\n"
+	}
+
+	headings := extractPDFHeadings(text)
+
+	if len(headings) > 10 {
+		t.Errorf("Expected max 10 headings, got %d", len(headings))
+	}
+}
+
+func TestContainsHeading(t *testing.T) {
+	headings := []string{"Title One", "Title Two", "Title Three"}
+
+	if !containsHeading(headings, "Title Two") {
+		t.Error("Expected to find 'Title Two'")
+	}
+
+	if containsHeading(headings, "Title Four") {
+		t.Error("Should not find 'Title Four'")
+	}
+
+	if containsHeading([]string{}, "Any") {
+		t.Error("Empty list should not contain anything")
+	}
+}
+
+func TestExtractPDFFallback_BasicExtraction(t *testing.T) {
+	// Test fallback with text in parentheses (PDF text stream format)
+	pdfLike := []byte("stream\n(Hello World) (This is some text) (More content here)\nendstream")
+
+	content, err := extractPDFFallback(pdfLike)
+	if err != nil {
+		t.Fatalf("extractPDFFallback failed: %v", err)
+	}
+
+	// Should extract text from parentheses
+	if !strings.Contains(content.ContentText, "Hello World") && content.ContentLength > 0 {
+		t.Log("Extracted some content via fallback")
+	}
+
+	// Should mark as fallback
+	if content.MetaData["extraction_method"] != "fallback" {
+		t.Error("Expected extraction_method to be 'fallback'")
+	}
+}
+
+func TestExtractPDF_EmptyInput(t *testing.T) {
+	content, err := ExtractPDF([]byte{})
+	if err != nil {
+		t.Fatalf("ExtractPDF should handle empty input: %v", err)
+	}
+
+	if content == nil {
+		t.Fatal("Expected non-nil content for empty input")
+	}
+
+	if content.ContentLength != 0 {
+		t.Errorf("Expected 0 content length for empty input, got %d", content.ContentLength)
+	}
+}
+
+func TestExtractPDFWithMetadata_FallbackOnError(t *testing.T) {
+	// ExtractPDFWithMetadata should fallback gracefully
+	content, err := ExtractPDFWithMetadata([]byte("not a pdf"))
+	if err != nil {
+		t.Fatalf("ExtractPDFWithMetadata should not fail: %v", err)
+	}
+
+	if content == nil {
+		t.Fatal("Expected non-nil content")
+	}
+}
+
+func TestExtractPDF_LanguageDetection(t *testing.T) {
+	tests := []struct {
+		name     string
+		text     string
+		expected string
+	}{
+		{
+			name:     "German content",
+			text:     "(Der Lehrplan ist für alle Schulen verbindlich und enthält wichtige Informationen)",
+			expected: "de",
+		},
+		{
+			name:     "Default to German",
+			text:     "(Some text)",
+			expected: "de",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			content, err := ExtractPDF([]byte(tt.text))
+			if err != nil {
+				t.Fatalf("ExtractPDF failed: %v", err)
+			}
+
+			// Language should be detected
+			if content.Language != tt.expected {
+				t.Logf("Language detected: %s (expected %s)", content.Language, tt.expected)
+			}
+		})
+	}
+}
--- a/edu-search-service/internal/indexer/mapping.go
+++ b/edu-search-service/internal/indexer/mapping.go
@@ -0,0 +1,243 @@
+package indexer
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"time"
+
+	"github.com/opensearch-project/opensearch-go/v2"
+	"github.com/opensearch-project/opensearch-go/v2/opensearchapi"
+)
+
+// IndexMapping defines the OpenSearch index mapping for education documents
+const IndexMapping = `{
+	"settings": {
+		"index": {
+			"number_of_shards": 3,
+			"number_of_replicas": 1,
+			"refresh_interval": "5s"
+		},
+		"analysis": {
+			"analyzer": {
+				"german_custom": {
+					"type": "custom",
+					"tokenizer": "standard",
+					"filter": ["lowercase", "german_normalization", "german_stemmer"]
+				}
+			},
+			"filter": {
+				"german_stemmer": {
+					"type": "stemmer",
+					"language": "german"
+				}
+			}
+		}
+	},
+	"mappings": {
+		"properties": {
+			"doc_id": { "type": "keyword" },
+			"url": { "type": "keyword" },
+			"canonical_url": { "type": "keyword" },
+			"domain": { "type": "keyword" },
+			"fetch_time": { "type": "date" },
+			"last_modified": { "type": "date" },
+			"content_hash": { "type": "keyword" },
+			"title": {
+				"type": "text",
+				"analyzer": "german_custom",
+				"fields": {
+					"keyword": { "type": "keyword", "ignore_above": 512 }
+				}
+			},
+			"content_text": {
+				"type": "text",
+				"analyzer": "german_custom"
+			},
+			"snippet_text": { "type": "text", "index": false },
+			"content_type": { "type": "keyword" },
+			"language": { "type": "keyword" },
+			"country_hint": { "type": "keyword" },
+			"source_category": { "type": "keyword" },
+			"doc_type": { "type": "keyword" },
+			"school_level": { "type": "keyword" },
+			"subjects": { "type": "keyword" },
+			"state": { "type": "keyword" },
+			"trust_score": { "type": "float" },
+			"quality_score": { "type": "float" },
+			"spam_flags": { "type": "keyword" },
+			"outlinks": { "type": "keyword" },
+			"inlinks_count": { "type": "integer" },
+			"content_length": { "type": "integer" },
+			"raw_refs": {
+				"properties": {
+					"html_raw_ref": { "type": "keyword" },
+					"pdf_raw_ref": { "type": "keyword" }
+				}
+			},
+			"tag_reasons": { "type": "keyword" }
+		}
+	}
+}`
+
+// Document represents an indexed education document
+type Document struct {
+	DocID          string    `json:"doc_id"`
+	URL            string    `json:"url"`
+	CanonicalURL   string    `json:"canonical_url,omitempty"`
+	Domain         string    `json:"domain"`
+	FetchedAt      time.Time `json:"fetch_time"`
+	UpdatedAt      time.Time `json:"last_modified,omitempty"`
+	ContentHash    string    `json:"content_hash"`
+	Title          string    `json:"title"`
+	ContentText    string    `json:"content_text"`
+	SnippetText    string    `json:"snippet_text"`
+	ContentType    string    `json:"content_type,omitempty"`
+	Language       string    `json:"language"`
+	CountryHint    string    `json:"country_hint,omitempty"`
+	SourceCategory string    `json:"source_category,omitempty"`
+	DocType        string    `json:"doc_type"`
+	SchoolLevel    string    `json:"school_level"`
+	Subjects       []string  `json:"subjects"`
+	State          string    `json:"state,omitempty"`
+	TrustScore     float64   `json:"trust_score"`
+	QualityScore   float64   `json:"quality_score"`
+	SpamFlags      []string  `json:"spam_flags,omitempty"`
+	Outlinks       []string  `json:"outlinks,omitempty"`
+	InlinksCount   int       `json:"inlinks_count,omitempty"`
+	ContentLength  int       `json:"content_length,omitempty"`
+	TagReasons     []string  `json:"tag_reasons,omitempty"`
+}
+
+// Client wraps OpenSearch operations
+type Client struct {
+	client    *opensearch.Client
+	indexName string
+}
+
+// NewClient creates a new OpenSearch indexer client
+func NewClient(url, username, password, indexName string) (*Client, error) {
+	cfg := opensearch.Config{
+		Addresses: []string{url},
+		Username:  username,
+		Password:  password,
+	}
+
+	client, err := opensearch.NewClient(cfg)
+	if err != nil {
+		return nil, err
+	}
+
+	return &Client{
+		client:    client,
+		indexName: indexName,
+	}, nil
+}
+
+// CreateIndex creates the index with proper mapping
+func (c *Client) CreateIndex(ctx context.Context) error {
+	// Check if index exists
+	res, err := c.client.Indices.Exists([]string{c.indexName})
+	if err != nil {
+		return err
+	}
+	defer res.Body.Close()
+
+	if res.StatusCode == 200 {
+		// Index already exists
+		return nil
+	}
+
+	// Create index with mapping
+	req := opensearchapi.IndicesCreateRequest{
+		Index: c.indexName,
+		Body:  strings.NewReader(IndexMapping),
+	}
+
+	res, err = req.Do(ctx, c.client)
+	if err != nil {
+		return err
+	}
+	defer res.Body.Close()
+
+	return nil
+}
+
+// IndexDocument indexes a single document
+func (c *Client) IndexDocument(ctx context.Context, doc *Document) error {
+	body, err := json.Marshal(doc)
+	if err != nil {
+		return err
+	}
+
+	req := opensearchapi.IndexRequest{
+		Index:      c.indexName,
+		DocumentID: doc.DocID,
+		Body:       strings.NewReader(string(body)),
+		Refresh:    "false",
+	}
+
+	res, err := req.Do(ctx, c.client)
+	if err != nil {
+		return err
+	}
+	defer res.Body.Close()
+
+	return nil
+}
+
+// BulkIndex indexes multiple documents efficiently
+func (c *Client) BulkIndex(ctx context.Context, docs []Document) error {
+	if len(docs) == 0 {
+		return nil
+	}
+
+	var builder strings.Builder
+
+	for _, doc := range docs {
+		// Action line
+		meta := map[string]interface{}{
+			"index": map[string]interface{}{
+				"_index": c.indexName,
+				"_id":    doc.DocID,
+			},
+		}
+		metaBytes, _ := json.Marshal(meta)
+		builder.Write(metaBytes)
+		builder.WriteString("\n")
+
+		// Document line
+		docBytes, _ := json.Marshal(doc)
+		builder.Write(docBytes)
+		builder.WriteString("\n")
+	}
+
+	req := opensearchapi.BulkRequest{
+		Body: strings.NewReader(builder.String()),
+	}
+
+	res, err := req.Do(ctx, c.client)
+	if err != nil {
+		return err
+	}
+	defer res.Body.Close()
+
+	return nil
+}
+
+// Health checks OpenSearch cluster health
+func (c *Client) Health(ctx context.Context) (string, error) {
+	res, err := c.client.Cluster.Health()
+	if err != nil {
+		return "", err
+	}
+	defer res.Body.Close()
+
+	var result map[string]interface{}
+	if err := json.NewDecoder(res.Body).Decode(&result); err != nil {
+		return "", err
+	}
+
+	status, _ := result["status"].(string)
+	return status, nil
+}
--- a/edu-search-service/internal/orchestrator/audiences.go
+++ b/edu-search-service/internal/orchestrator/audiences.go
@@ -0,0 +1,424 @@
+// Package orchestrator implements multi-phase university crawling with queue management
+package orchestrator
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"time"
+
+	"github.com/google/uuid"
+)
+
+// Audience represents a target audience filter configuration
+type Audience struct {
+	ID              uuid.UUID        `json:"id"`
+	Name            string           `json:"name"`
+	Description     string           `json:"description,omitempty"`
+	Filters         AudienceFilters  `json:"filters"`
+	MemberCount     int              `json:"member_count"`
+	LastCountUpdate *time.Time       `json:"last_count_update,omitempty"`
+	CreatedBy       string           `json:"created_by,omitempty"`
+	IsActive        bool             `json:"is_active"`
+	CreatedAt       time.Time        `json:"created_at"`
+	UpdatedAt       time.Time        `json:"updated_at"`
+}
+
+// AudienceFilters defines the filter criteria for an audience
+type AudienceFilters struct {
+	PositionTypes []string    `json:"position_types,omitempty"` // professor, researcher, lecturer
+	SubjectAreas  []uuid.UUID `json:"subject_areas,omitempty"`  // Subject area UUIDs
+	States        []string    `json:"states,omitempty"`         // BW, BY, etc.
+	UniTypes      []string    `json:"uni_types,omitempty"`      // UNI, PH, HAW
+	Universities  []uuid.UUID `json:"universities,omitempty"`   // University UUIDs
+	HasEmail      *bool       `json:"has_email,omitempty"`
+	IsActive      *bool       `json:"is_active,omitempty"`
+	Keywords      []string    `json:"keywords,omitempty"` // Keywords in name/research
+}
+
+// AudienceExport tracks exports of audience data
+type AudienceExport struct {
+	ID          uuid.UUID  `json:"id"`
+	AudienceID  uuid.UUID  `json:"audience_id"`
+	ExportType  string     `json:"export_type"` // csv, json, email_list
+	RecordCount int        `json:"record_count"`
+	FilePath    string     `json:"file_path,omitempty"`
+	ExportedBy  string     `json:"exported_by,omitempty"`
+	Purpose     string     `json:"purpose,omitempty"`
+	CreatedAt   time.Time  `json:"created_at"`
+}
+
+// AudienceMember represents a staff member in an audience preview
+type AudienceMember struct {
+	ID             uuid.UUID `json:"id"`
+	Name           string    `json:"name"`
+	Email          string    `json:"email,omitempty"`
+	Position       string    `json:"position,omitempty"`
+	University     string    `json:"university"`
+	Department     string    `json:"department,omitempty"`
+	SubjectArea    string    `json:"subject_area,omitempty"`
+	PublicationCount int     `json:"publication_count"`
+}
+
+// AudienceRepository extends Repository with audience operations
+type AudienceRepository interface {
+	// Audience CRUD
+	CreateAudience(ctx context.Context, audience *Audience) error
+	GetAudience(ctx context.Context, id uuid.UUID) (*Audience, error)
+	ListAudiences(ctx context.Context, activeOnly bool) ([]Audience, error)
+	UpdateAudience(ctx context.Context, audience *Audience) error
+	DeleteAudience(ctx context.Context, id uuid.UUID) error
+
+	// Audience members
+	GetAudienceMembers(ctx context.Context, id uuid.UUID, limit, offset int) ([]AudienceMember, int, error)
+	UpdateAudienceCount(ctx context.Context, id uuid.UUID) (int, error)
+
+	// Exports
+	CreateExport(ctx context.Context, export *AudienceExport) error
+	ListExports(ctx context.Context, audienceID uuid.UUID) ([]AudienceExport, error)
+}
+
+// ============================================================================
+// POSTGRES IMPLEMENTATION
+// ============================================================================
+
+// CreateAudience creates a new audience
+func (r *PostgresRepository) CreateAudience(ctx context.Context, audience *Audience) error {
+	filtersJSON, err := json.Marshal(audience.Filters)
+	if err != nil {
+		return fmt.Errorf("failed to marshal filters: %w", err)
+	}
+
+	query := `
+		INSERT INTO audiences (name, description, filters, created_by, is_active)
+		VALUES ($1, $2, $3, $4, $5)
+		RETURNING id, member_count, created_at, updated_at
+	`
+
+	return r.pool.QueryRow(ctx, query,
+		audience.Name,
+		audience.Description,
+		filtersJSON,
+		audience.CreatedBy,
+		audience.IsActive,
+	).Scan(&audience.ID, &audience.MemberCount, &audience.CreatedAt, &audience.UpdatedAt)
+}
+
+// GetAudience retrieves an audience by ID
+func (r *PostgresRepository) GetAudience(ctx context.Context, id uuid.UUID) (*Audience, error) {
+	query := `
+		SELECT id, name, description, filters, member_count, last_count_update,
+		       created_by, is_active, created_at, updated_at
+		FROM audiences
+		WHERE id = $1
+	`
+
+	var audience Audience
+	var filtersJSON []byte
+
+	err := r.pool.QueryRow(ctx, query, id).Scan(
+		&audience.ID, &audience.Name, &audience.Description, &filtersJSON,
+		&audience.MemberCount, &audience.LastCountUpdate,
+		&audience.CreatedBy, &audience.IsActive,
+		&audience.CreatedAt, &audience.UpdatedAt,
+	)
+	if err != nil {
+		return nil, err
+	}
+
+	if err := json.Unmarshal(filtersJSON, &audience.Filters); err != nil {
+		return nil, fmt.Errorf("failed to unmarshal filters: %w", err)
+	}
+
+	return &audience, nil
+}
+
+// ListAudiences lists all audiences
+func (r *PostgresRepository) ListAudiences(ctx context.Context, activeOnly bool) ([]Audience, error) {
+	query := `
+		SELECT id, name, description, filters, member_count, last_count_update,
+		       created_by, is_active, created_at, updated_at
+		FROM audiences
+	`
+	if activeOnly {
+		query += ` WHERE is_active = TRUE`
+	}
+	query += ` ORDER BY created_at DESC`
+
+	rows, err := r.pool.Query(ctx, query)
+	if err != nil {
+		return nil, fmt.Errorf("failed to query audiences: %w", err)
+	}
+	defer rows.Close()
+
+	var audiences []Audience
+	for rows.Next() {
+		var audience Audience
+		var filtersJSON []byte
+
+		if err := rows.Scan(
+			&audience.ID, &audience.Name, &audience.Description, &filtersJSON,
+			&audience.MemberCount, &audience.LastCountUpdate,
+			&audience.CreatedBy, &audience.IsActive,
+			&audience.CreatedAt, &audience.UpdatedAt,
+		); err != nil {
+			return nil, fmt.Errorf("failed to scan audience: %w", err)
+		}
+
+		if err := json.Unmarshal(filtersJSON, &audience.Filters); err != nil {
+			return nil, fmt.Errorf("failed to unmarshal filters: %w", err)
+		}
+
+		audiences = append(audiences, audience)
+	}
+
+	return audiences, rows.Err()
+}
+
+// UpdateAudience updates an existing audience
+func (r *PostgresRepository) UpdateAudience(ctx context.Context, audience *Audience) error {
+	filtersJSON, err := json.Marshal(audience.Filters)
+	if err != nil {
+		return fmt.Errorf("failed to marshal filters: %w", err)
+	}
+
+	query := `
+		UPDATE audiences
+		SET name = $2, description = $3, filters = $4, is_active = $5, updated_at = NOW()
+		WHERE id = $1
+		RETURNING updated_at
+	`
+
+	return r.pool.QueryRow(ctx, query,
+		audience.ID,
+		audience.Name,
+		audience.Description,
+		filtersJSON,
+		audience.IsActive,
+	).Scan(&audience.UpdatedAt)
+}
+
+// DeleteAudience soft-deletes an audience (sets is_active = false)
+func (r *PostgresRepository) DeleteAudience(ctx context.Context, id uuid.UUID) error {
+	query := `UPDATE audiences SET is_active = FALSE, updated_at = NOW() WHERE id = $1`
+	_, err := r.pool.Exec(ctx, query, id)
+	return err
+}
+
+// GetAudienceMembers retrieves members matching the audience filters
+func (r *PostgresRepository) GetAudienceMembers(ctx context.Context, id uuid.UUID, limit, offset int) ([]AudienceMember, int, error) {
+	// First get the audience filters
+	audience, err := r.GetAudience(ctx, id)
+	if err != nil {
+		return nil, 0, fmt.Errorf("failed to get audience: %w", err)
+	}
+
+	// Build dynamic query based on filters
+	query, args := r.buildAudienceMemberQuery(audience.Filters, limit, offset, false)
+	countQuery, countArgs := r.buildAudienceMemberQuery(audience.Filters, 0, 0, true)
+
+	// Get total count
+	var totalCount int
+	if err := r.pool.QueryRow(ctx, countQuery, countArgs...).Scan(&totalCount); err != nil {
+		return nil, 0, fmt.Errorf("failed to count members: %w", err)
+	}
+
+	// Get members
+	rows, err := r.pool.Query(ctx, query, args...)
+	if err != nil {
+		return nil, 0, fmt.Errorf("failed to query members: %w", err)
+	}
+	defer rows.Close()
+
+	var members []AudienceMember
+	for rows.Next() {
+		var m AudienceMember
+		if err := rows.Scan(
+			&m.ID, &m.Name, &m.Email, &m.Position,
+			&m.University, &m.Department, &m.SubjectArea, &m.PublicationCount,
+		); err != nil {
+			return nil, 0, fmt.Errorf("failed to scan member: %w", err)
+		}
+		members = append(members, m)
+	}
+
+	return members, totalCount, rows.Err()
+}
+
+// buildAudienceMemberQuery constructs a SQL query for audience members
+func (r *PostgresRepository) buildAudienceMemberQuery(filters AudienceFilters, limit, offset int, countOnly bool) (string, []interface{}) {
+	var args []interface{}
+	argNum := 1
+
+	var selectClause string
+	if countOnly {
+		selectClause = "SELECT COUNT(*)"
+	} else {
+		selectClause = `
+			SELECT
+				s.id,
+				COALESCE(s.title || ' ', '') || s.first_name || ' ' || s.last_name as name,
+				COALESCE(s.email, '') as email,
+				COALESCE(s.position_type, '') as position,
+				u.name as university,
+				COALESCE(d.name, '') as department,
+				COALESCE(sa.name, '') as subject_area,
+				(SELECT COUNT(*) FROM staff_publications sp WHERE sp.staff_id = s.id) as publication_count
+		`
+	}
+
+	query := selectClause + `
+		FROM university_staff s
+		JOIN universities u ON s.university_id = u.id
+		LEFT JOIN departments d ON s.department_id = d.id
+		LEFT JOIN subject_areas sa ON s.subject_area_id = sa.id
+		WHERE 1=1
+	`
+
+	// Position types filter
+	if len(filters.PositionTypes) > 0 {
+		query += fmt.Sprintf(" AND s.position_type = ANY($%d)", argNum)
+		args = append(args, filters.PositionTypes)
+		argNum++
+	}
+
+	// Subject areas filter
+	if len(filters.SubjectAreas) > 0 {
+		query += fmt.Sprintf(" AND s.subject_area_id = ANY($%d)", argNum)
+		args = append(args, filters.SubjectAreas)
+		argNum++
+	}
+
+	// States filter
+	if len(filters.States) > 0 {
+		query += fmt.Sprintf(" AND u.state = ANY($%d)", argNum)
+		args = append(args, filters.States)
+		argNum++
+	}
+
+	// Uni types filter
+	if len(filters.UniTypes) > 0 {
+		query += fmt.Sprintf(" AND u.uni_type = ANY($%d)", argNum)
+		args = append(args, filters.UniTypes)
+		argNum++
+	}
+
+	// Universities filter
+	if len(filters.Universities) > 0 {
+		query += fmt.Sprintf(" AND s.university_id = ANY($%d)", argNum)
+		args = append(args, filters.Universities)
+		argNum++
+	}
+
+	// Has email filter
+	if filters.HasEmail != nil && *filters.HasEmail {
+		query += " AND s.email IS NOT NULL AND s.email != ''"
+	}
+
+	// Is active filter
+	if filters.IsActive != nil && *filters.IsActive {
+		query += " AND s.is_active = TRUE"
+	}
+
+	// Keywords filter (search in name and research_areas)
+	if len(filters.Keywords) > 0 {
+		for _, keyword := range filters.Keywords {
+			query += fmt.Sprintf(" AND (s.first_name ILIKE $%d OR s.last_name ILIKE $%d OR s.research_areas ILIKE $%d)", argNum, argNum, argNum)
+			args = append(args, "%"+keyword+"%")
+			argNum++
+		}
+	}
+
+	if !countOnly {
+		query += " ORDER BY s.last_name, s.first_name"
+
+		if limit > 0 {
+			query += fmt.Sprintf(" LIMIT $%d", argNum)
+			args = append(args, limit)
+			argNum++
+		}
+
+		if offset > 0 {
+			query += fmt.Sprintf(" OFFSET $%d", argNum)
+			args = append(args, offset)
+		}
+	}
+
+	return query, args
+}
+
+// UpdateAudienceCount updates the cached member count for an audience
+func (r *PostgresRepository) UpdateAudienceCount(ctx context.Context, id uuid.UUID) (int, error) {
+	// Get the audience filters
+	audience, err := r.GetAudience(ctx, id)
+	if err != nil {
+		return 0, fmt.Errorf("failed to get audience: %w", err)
+	}
+
+	// Count members
+	countQuery, countArgs := r.buildAudienceMemberQuery(audience.Filters, 0, 0, true)
+	var count int
+	if err := r.pool.QueryRow(ctx, countQuery, countArgs...).Scan(&count); err != nil {
+		return 0, fmt.Errorf("failed to count members: %w", err)
+	}
+
+	// Update the cached count
+	updateQuery := `
+		UPDATE audiences
+		SET member_count = $2, last_count_update = NOW(), updated_at = NOW()
+		WHERE id = $1
+	`
+	if _, err := r.pool.Exec(ctx, updateQuery, id, count); err != nil {
+		return 0, fmt.Errorf("failed to update count: %w", err)
+	}
+
+	return count, nil
+}
+
+// CreateExport creates a new export record
+func (r *PostgresRepository) CreateExport(ctx context.Context, export *AudienceExport) error {
+	query := `
+		INSERT INTO audience_exports (audience_id, export_type, record_count, file_path, exported_by, purpose)
+		VALUES ($1, $2, $3, $4, $5, $6)
+		RETURNING id, created_at
+	`
+
+	return r.pool.QueryRow(ctx, query,
+		export.AudienceID,
+		export.ExportType,
+		export.RecordCount,
+		export.FilePath,
+		export.ExportedBy,
+		export.Purpose,
+	).Scan(&export.ID, &export.CreatedAt)
+}
+
+// ListExports lists exports for an audience
+func (r *PostgresRepository) ListExports(ctx context.Context, audienceID uuid.UUID) ([]AudienceExport, error) {
+	query := `
+		SELECT id, audience_id, export_type, record_count, file_path, exported_by, purpose, created_at
+		FROM audience_exports
+		WHERE audience_id = $1
+		ORDER BY created_at DESC
+	`
+
+	rows, err := r.pool.Query(ctx, query, audienceID)
+	if err != nil {
+		return nil, fmt.Errorf("failed to query exports: %w", err)
+	}
+	defer rows.Close()
+
+	var exports []AudienceExport
+	for rows.Next() {
+		var e AudienceExport
+		if err := rows.Scan(
+			&e.ID, &e.AudienceID, &e.ExportType, &e.RecordCount,
+			&e.FilePath, &e.ExportedBy, &e.Purpose, &e.CreatedAt,
+		); err != nil {
+			return nil, fmt.Errorf("failed to scan export: %w", err)
+		}
+		exports = append(exports, e)
+	}
+
+	return exports, rows.Err()
+}
--- a/edu-search-service/internal/orchestrator/orchestrator.go
+++ b/edu-search-service/internal/orchestrator/orchestrator.go
@@ -0,0 +1,407 @@
+// Package orchestrator implements multi-phase university crawling with queue management
+package orchestrator
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"sync"
+	"time"
+
+	"github.com/google/uuid"
+)
+
+// CrawlPhase represents a phase in the crawl process
+type CrawlPhase string
+
+const (
+	PhasePending      CrawlPhase = "pending"
+	PhaseDiscovery    CrawlPhase = "discovery"    // Find sample professor to validate crawling works
+	PhaseProfessors   CrawlPhase = "professors"   // Crawl all professors
+	PhaseAllStaff     CrawlPhase = "all_staff"    // Crawl all staff members
+	PhasePublications CrawlPhase = "publications" // Crawl publications for all staff
+	PhaseCompleted    CrawlPhase = "completed"
+	PhaseFailed       CrawlPhase = "failed"
+	PhasePaused       CrawlPhase = "paused"
+)
+
+// CrawlQueueItem represents a university in the crawl queue
+type CrawlQueueItem struct {
+	ID                     uuid.UUID  `json:"id"`
+	UniversityID           uuid.UUID  `json:"university_id"`
+	UniversityName         string     `json:"university_name"`
+	UniversityShort        string     `json:"university_short"`
+	QueuePosition          *int       `json:"queue_position"`
+	Priority               int        `json:"priority"`
+	CurrentPhase           CrawlPhase `json:"current_phase"`
+	DiscoveryCompleted     bool       `json:"discovery_completed"`
+	DiscoveryCompletedAt   *time.Time `json:"discovery_completed_at,omitempty"`
+	ProfessorsCompleted    bool       `json:"professors_completed"`
+	ProfessorsCompletedAt  *time.Time `json:"professors_completed_at,omitempty"`
+	AllStaffCompleted      bool       `json:"all_staff_completed"`
+	AllStaffCompletedAt    *time.Time `json:"all_staff_completed_at,omitempty"`
+	PublicationsCompleted  bool       `json:"publications_completed"`
+	PublicationsCompletedAt *time.Time `json:"publications_completed_at,omitempty"`
+	DiscoveryCount         int        `json:"discovery_count"`
+	ProfessorsCount        int        `json:"professors_count"`
+	StaffCount             int        `json:"staff_count"`
+	PublicationsCount      int        `json:"publications_count"`
+	RetryCount             int        `json:"retry_count"`
+	MaxRetries             int        `json:"max_retries"`
+	LastError              string     `json:"last_error,omitempty"`
+	StartedAt              *time.Time `json:"started_at,omitempty"`
+	CompletedAt            *time.Time `json:"completed_at,omitempty"`
+	ProgressPercent        int        `json:"progress_percent"`
+	CreatedAt              time.Time  `json:"created_at"`
+	UpdatedAt              time.Time  `json:"updated_at"`
+}
+
+// CrawlProgress represents progress for a single phase
+type CrawlProgress struct {
+	Phase          CrawlPhase `json:"phase"`
+	ItemsFound     int        `json:"items_found"`
+	ItemsProcessed int        `json:"items_processed"`
+	Errors         []string   `json:"errors,omitempty"`
+	StartedAt      time.Time  `json:"started_at"`
+	CompletedAt    *time.Time `json:"completed_at,omitempty"`
+}
+
+// OrchestratorStatus represents the current state of the orchestrator
+type OrchestratorStatus struct {
+	IsRunning         bool             `json:"is_running"`
+	CurrentUniversity *CrawlQueueItem  `json:"current_university,omitempty"`
+	CurrentPhase      CrawlPhase       `json:"current_phase"`
+	QueueLength       int              `json:"queue_length"`
+	CompletedToday    int              `json:"completed_today"`
+	TotalProcessed    int              `json:"total_processed"`
+	LastActivity      *time.Time       `json:"last_activity,omitempty"`
+}
+
+// StaffCrawlerInterface defines what the staff crawler must implement
+type StaffCrawlerInterface interface {
+	// DiscoverSampleProfessor finds at least one professor to validate crawling works
+	DiscoverSampleProfessor(ctx context.Context, universityID uuid.UUID) (*CrawlProgress, error)
+	// CrawlProfessors crawls all professors at a university
+	CrawlProfessors(ctx context.Context, universityID uuid.UUID) (*CrawlProgress, error)
+	// CrawlAllStaff crawls all staff members at a university
+	CrawlAllStaff(ctx context.Context, universityID uuid.UUID) (*CrawlProgress, error)
+}
+
+// PublicationCrawlerInterface defines what the publication crawler must implement
+type PublicationCrawlerInterface interface {
+	// CrawlPublicationsForUniversity crawls publications for all staff at a university
+	CrawlPublicationsForUniversity(ctx context.Context, universityID uuid.UUID) (*CrawlProgress, error)
+}
+
+// Repository defines database operations for the orchestrator
+type Repository interface {
+	// Queue operations
+	GetQueueItems(ctx context.Context) ([]CrawlQueueItem, error)
+	GetNextInQueue(ctx context.Context) (*CrawlQueueItem, error)
+	AddToQueue(ctx context.Context, universityID uuid.UUID, priority int, initiatedBy string) (*CrawlQueueItem, error)
+	RemoveFromQueue(ctx context.Context, universityID uuid.UUID) error
+	UpdateQueueItem(ctx context.Context, item *CrawlQueueItem) error
+	PauseQueueItem(ctx context.Context, universityID uuid.UUID) error
+	ResumeQueueItem(ctx context.Context, universityID uuid.UUID) error
+
+	// Phase updates
+	CompletePhase(ctx context.Context, universityID uuid.UUID, phase CrawlPhase, count int) error
+	FailPhase(ctx context.Context, universityID uuid.UUID, phase CrawlPhase, err string) error
+
+	// Stats
+	GetCompletedTodayCount(ctx context.Context) (int, error)
+	GetTotalProcessedCount(ctx context.Context) (int, error)
+}
+
+// Orchestrator manages the multi-phase crawl process
+type Orchestrator struct {
+	repo         Repository
+	staffCrawler StaffCrawlerInterface
+	pubCrawler   PublicationCrawlerInterface
+
+	// Runtime state
+	mu            sync.RWMutex
+	isRunning     bool
+	stopChan      chan struct{}
+	currentItem   *CrawlQueueItem
+	lastActivity  time.Time
+
+	// Configuration
+	phaseCooldown  time.Duration // Wait time between phases
+	retryCooldown  time.Duration // Wait time after failure before retry
+	maxConcurrent  int           // Max concurrent crawls (always 1 for now)
+}
+
+// NewOrchestrator creates a new orchestrator instance
+func NewOrchestrator(repo Repository, staffCrawler StaffCrawlerInterface, pubCrawler PublicationCrawlerInterface) *Orchestrator {
+	return &Orchestrator{
+		repo:          repo,
+		staffCrawler:  staffCrawler,
+		pubCrawler:    pubCrawler,
+		phaseCooldown: 5 * time.Second,  // Small pause between phases
+		retryCooldown: 30 * time.Second, // Wait before retry after failure
+		maxConcurrent: 1,                // Sequential processing
+	}
+}
+
+// Start begins the orchestrator loop
+func (o *Orchestrator) Start() error {
+	o.mu.Lock()
+	if o.isRunning {
+		o.mu.Unlock()
+		return fmt.Errorf("orchestrator already running")
+	}
+	o.isRunning = true
+	o.stopChan = make(chan struct{})
+	o.mu.Unlock()
+
+	log.Println("[Orchestrator] Starting crawl orchestration loop")
+
+	go o.runLoop()
+	return nil
+}
+
+// Stop gracefully stops the orchestrator
+func (o *Orchestrator) Stop() error {
+	o.mu.Lock()
+	if !o.isRunning {
+		o.mu.Unlock()
+		return fmt.Errorf("orchestrator not running")
+	}
+	close(o.stopChan)
+	o.isRunning = false
+	o.mu.Unlock()
+
+	log.Println("[Orchestrator] Stopped")
+	return nil
+}
+
+// Status returns the current orchestrator status
+func (o *Orchestrator) Status(ctx context.Context) (*OrchestratorStatus, error) {
+	o.mu.RLock()
+	defer o.mu.RUnlock()
+
+	status := &OrchestratorStatus{
+		IsRunning:    o.isRunning,
+		CurrentPhase: PhasePending,
+	}
+
+	if o.currentItem != nil {
+		status.CurrentUniversity = o.currentItem
+		status.CurrentPhase = o.currentItem.CurrentPhase
+	}
+
+	if !o.lastActivity.IsZero() {
+		status.LastActivity = &o.lastActivity
+	}
+
+	// Get queue stats from DB
+	items, err := o.repo.GetQueueItems(ctx)
+	if err == nil {
+		status.QueueLength = len(items)
+	}
+
+	completedToday, _ := o.repo.GetCompletedTodayCount(ctx)
+	status.CompletedToday = completedToday
+
+	totalProcessed, _ := o.repo.GetTotalProcessedCount(ctx)
+	status.TotalProcessed = totalProcessed
+
+	return status, nil
+}
+
+// AddUniversity adds a university to the crawl queue
+func (o *Orchestrator) AddUniversity(ctx context.Context, universityID uuid.UUID, priority int, initiatedBy string) (*CrawlQueueItem, error) {
+	item, err := o.repo.AddToQueue(ctx, universityID, priority, initiatedBy)
+	if err != nil {
+		return nil, fmt.Errorf("failed to add to queue: %w", err)
+	}
+
+	log.Printf("[Orchestrator] Added university %s to queue with priority %d", universityID, priority)
+	return item, nil
+}
+
+// RemoveUniversity removes a university from the queue
+func (o *Orchestrator) RemoveUniversity(ctx context.Context, universityID uuid.UUID) error {
+	return o.repo.RemoveFromQueue(ctx, universityID)
+}
+
+// PauseUniversity pauses crawling for a university
+func (o *Orchestrator) PauseUniversity(ctx context.Context, universityID uuid.UUID) error {
+	return o.repo.PauseQueueItem(ctx, universityID)
+}
+
+// ResumeUniversity resumes crawling for a paused university
+func (o *Orchestrator) ResumeUniversity(ctx context.Context, universityID uuid.UUID) error {
+	return o.repo.ResumeQueueItem(ctx, universityID)
+}
+
+// GetQueue returns all items in the queue
+func (o *Orchestrator) GetQueue(ctx context.Context) ([]CrawlQueueItem, error) {
+	return o.repo.GetQueueItems(ctx)
+}
+
+// runLoop is the main orchestration loop
+func (o *Orchestrator) runLoop() {
+	ticker := time.NewTicker(10 * time.Second) // Check queue every 10 seconds
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-o.stopChan:
+			return
+		case <-ticker.C:
+			o.processNextInQueue()
+		}
+	}
+}
+
+// processNextInQueue processes the next university in the queue
+func (o *Orchestrator) processNextInQueue() {
+	ctx := context.Background()
+
+	// Get next item in queue
+	item, err := o.repo.GetNextInQueue(ctx)
+	if err != nil {
+		log.Printf("[Orchestrator] Error getting next item: %v", err)
+		return
+	}
+
+	if item == nil {
+		// No items to process
+		return
+	}
+
+	// Check if paused
+	if item.CurrentPhase == PhasePaused {
+		return
+	}
+
+	// Set current item
+	o.mu.Lock()
+	o.currentItem = item
+	o.lastActivity = time.Now()
+	o.mu.Unlock()
+
+	defer func() {
+		o.mu.Lock()
+		o.currentItem = nil
+		o.mu.Unlock()
+	}()
+
+	log.Printf("[Orchestrator] Processing university: %s (Phase: %s)", item.UniversityName, item.CurrentPhase)
+
+	// Process based on current phase
+	switch item.CurrentPhase {
+	case PhasePending:
+		o.runPhase(ctx, item, PhaseDiscovery)
+	case PhaseDiscovery:
+		if item.DiscoveryCompleted {
+			o.runPhase(ctx, item, PhaseProfessors)
+		} else {
+			o.runPhase(ctx, item, PhaseDiscovery)
+		}
+	case PhaseProfessors:
+		if item.ProfessorsCompleted {
+			o.runPhase(ctx, item, PhaseAllStaff)
+		} else {
+			o.runPhase(ctx, item, PhaseProfessors)
+		}
+	case PhaseAllStaff:
+		if item.AllStaffCompleted {
+			o.runPhase(ctx, item, PhasePublications)
+		} else {
+			o.runPhase(ctx, item, PhaseAllStaff)
+		}
+	case PhasePublications:
+		if item.PublicationsCompleted {
+			o.completeUniversity(ctx, item)
+		} else {
+			o.runPhase(ctx, item, PhasePublications)
+		}
+	}
+}
+
+// runPhase executes a specific crawl phase
+func (o *Orchestrator) runPhase(ctx context.Context, item *CrawlQueueItem, phase CrawlPhase) {
+	log.Printf("[Orchestrator] Running phase %s for %s", phase, item.UniversityName)
+
+	// Update current phase
+	item.CurrentPhase = phase
+	if err := o.repo.UpdateQueueItem(ctx, item); err != nil {
+		log.Printf("[Orchestrator] Failed to update phase: %v", err)
+		return
+	}
+
+	var progress *CrawlProgress
+	var err error
+
+	// Execute phase
+	switch phase {
+	case PhaseDiscovery:
+		progress, err = o.staffCrawler.DiscoverSampleProfessor(ctx, item.UniversityID)
+	case PhaseProfessors:
+		progress, err = o.staffCrawler.CrawlProfessors(ctx, item.UniversityID)
+	case PhaseAllStaff:
+		progress, err = o.staffCrawler.CrawlAllStaff(ctx, item.UniversityID)
+	case PhasePublications:
+		progress, err = o.pubCrawler.CrawlPublicationsForUniversity(ctx, item.UniversityID)
+	}
+
+	// Handle result
+	if err != nil {
+		log.Printf("[Orchestrator] Phase %s failed: %v", phase, err)
+		o.handlePhaseFailure(ctx, item, phase, err)
+		return
+	}
+
+	// Mark phase complete
+	count := 0
+	if progress != nil {
+		count = progress.ItemsFound
+	}
+
+	if err := o.repo.CompletePhase(ctx, item.UniversityID, phase, count); err != nil {
+		log.Printf("[Orchestrator] Failed to complete phase: %v", err)
+	}
+
+	log.Printf("[Orchestrator] Phase %s completed for %s (found: %d)", phase, item.UniversityName, count)
+
+	// Wait before next phase
+	time.Sleep(o.phaseCooldown)
+}
+
+// handlePhaseFailure handles a phase failure
+func (o *Orchestrator) handlePhaseFailure(ctx context.Context, item *CrawlQueueItem, phase CrawlPhase, err error) {
+	item.RetryCount++
+	item.LastError = err.Error()
+
+	if item.RetryCount >= item.MaxRetries {
+		// Max retries reached, mark as failed
+		item.CurrentPhase = PhaseFailed
+		log.Printf("[Orchestrator] University %s failed after %d retries", item.UniversityName, item.RetryCount)
+	}
+
+	if updateErr := o.repo.FailPhase(ctx, item.UniversityID, phase, err.Error()); updateErr != nil {
+		log.Printf("[Orchestrator] Failed to update failure status: %v", updateErr)
+	}
+
+	// Wait before potential retry
+	time.Sleep(o.retryCooldown)
+}
+
+// completeUniversity marks a university as fully crawled
+func (o *Orchestrator) completeUniversity(ctx context.Context, item *CrawlQueueItem) {
+	now := time.Now()
+	item.CurrentPhase = PhaseCompleted
+	item.CompletedAt = &now
+	item.QueuePosition = nil // Remove from active queue
+
+	if err := o.repo.UpdateQueueItem(ctx, item); err != nil {
+		log.Printf("[Orchestrator] Failed to complete university: %v", err)
+		return
+	}
+
+	log.Printf("[Orchestrator] University %s completed! Professors: %d, Staff: %d, Publications: %d",
+		item.UniversityName, item.ProfessorsCount, item.StaffCount, item.PublicationsCount)
+}
--- a/edu-search-service/internal/orchestrator/repository.go
+++ b/edu-search-service/internal/orchestrator/repository.go
@@ -0,0 +1,316 @@
+// Package orchestrator implements multi-phase university crawling with queue management
+package orchestrator
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/google/uuid"
+	"github.com/jackc/pgx/v5"
+	"github.com/jackc/pgx/v5/pgxpool"
+)
+
+// PostgresRepository implements the Repository interface using PostgreSQL
+type PostgresRepository struct {
+	pool *pgxpool.Pool
+}
+
+// NewPostgresRepository creates a new PostgresRepository
+func NewPostgresRepository(pool *pgxpool.Pool) *PostgresRepository {
+	return &PostgresRepository{pool: pool}
+}
+
+// ============================================================================
+// QUEUE OPERATIONS
+// ============================================================================
+
+// GetQueueItems retrieves all items in the crawl queue
+func (r *PostgresRepository) GetQueueItems(ctx context.Context) ([]CrawlQueueItem, error) {
+	query := `
+		SELECT
+			cq.id, cq.university_id, u.name, COALESCE(u.short_name, ''),
+			cq.queue_position, cq.priority, cq.current_phase,
+			cq.discovery_completed, cq.discovery_completed_at,
+			cq.professors_completed, cq.professors_completed_at,
+			cq.all_staff_completed, cq.all_staff_completed_at,
+			cq.publications_completed, cq.publications_completed_at,
+			cq.discovery_count, cq.professors_count, cq.staff_count, cq.publications_count,
+			cq.retry_count, cq.max_retries, COALESCE(cq.last_error, ''),
+			cq.started_at, cq.completed_at,
+			CASE
+				WHEN cq.current_phase = 'pending' THEN 0
+				WHEN cq.current_phase = 'discovery' THEN 10
+				WHEN cq.current_phase = 'professors' THEN 30
+				WHEN cq.current_phase = 'all_staff' THEN 60
+				WHEN cq.current_phase = 'publications' THEN 90
+				WHEN cq.current_phase = 'completed' THEN 100
+				ELSE 0
+			END as progress_percent,
+			cq.created_at, cq.updated_at
+		FROM crawl_queue cq
+		JOIN universities u ON cq.university_id = u.id
+		ORDER BY cq.queue_position NULLS LAST, cq.priority DESC
+	`
+
+	rows, err := r.pool.Query(ctx, query)
+	if err != nil {
+		return nil, fmt.Errorf("failed to query queue items: %w", err)
+	}
+	defer rows.Close()
+
+	var items []CrawlQueueItem
+	for rows.Next() {
+		var item CrawlQueueItem
+		var phase string
+		if err := rows.Scan(
+			&item.ID, &item.UniversityID, &item.UniversityName, &item.UniversityShort,
+			&item.QueuePosition, &item.Priority, &phase,
+			&item.DiscoveryCompleted, &item.DiscoveryCompletedAt,
+			&item.ProfessorsCompleted, &item.ProfessorsCompletedAt,
+			&item.AllStaffCompleted, &item.AllStaffCompletedAt,
+			&item.PublicationsCompleted, &item.PublicationsCompletedAt,
+			&item.DiscoveryCount, &item.ProfessorsCount, &item.StaffCount, &item.PublicationsCount,
+			&item.RetryCount, &item.MaxRetries, &item.LastError,
+			&item.StartedAt, &item.CompletedAt,
+			&item.ProgressPercent,
+			&item.CreatedAt, &item.UpdatedAt,
+		); err != nil {
+			return nil, fmt.Errorf("failed to scan queue item: %w", err)
+		}
+		item.CurrentPhase = CrawlPhase(phase)
+		items = append(items, item)
+	}
+
+	return items, rows.Err()
+}
+
+// GetNextInQueue retrieves the next item to process
+func (r *PostgresRepository) GetNextInQueue(ctx context.Context) (*CrawlQueueItem, error) {
+	query := `
+		SELECT
+			cq.id, cq.university_id, u.name, COALESCE(u.short_name, ''),
+			cq.queue_position, cq.priority, cq.current_phase,
+			cq.discovery_completed, cq.discovery_completed_at,
+			cq.professors_completed, cq.professors_completed_at,
+			cq.all_staff_completed, cq.all_staff_completed_at,
+			cq.publications_completed, cq.publications_completed_at,
+			cq.discovery_count, cq.professors_count, cq.staff_count, cq.publications_count,
+			cq.retry_count, cq.max_retries, COALESCE(cq.last_error, ''),
+			cq.started_at, cq.completed_at,
+			cq.created_at, cq.updated_at
+		FROM crawl_queue cq
+		JOIN universities u ON cq.university_id = u.id
+		WHERE cq.current_phase NOT IN ('completed', 'failed', 'paused')
+			AND cq.queue_position IS NOT NULL
+		ORDER BY cq.queue_position ASC, cq.priority DESC
+		LIMIT 1
+	`
+
+	var item CrawlQueueItem
+	var phase string
+	err := r.pool.QueryRow(ctx, query).Scan(
+		&item.ID, &item.UniversityID, &item.UniversityName, &item.UniversityShort,
+		&item.QueuePosition, &item.Priority, &phase,
+		&item.DiscoveryCompleted, &item.DiscoveryCompletedAt,
+		&item.ProfessorsCompleted, &item.ProfessorsCompletedAt,
+		&item.AllStaffCompleted, &item.AllStaffCompletedAt,
+		&item.PublicationsCompleted, &item.PublicationsCompletedAt,
+		&item.DiscoveryCount, &item.ProfessorsCount, &item.StaffCount, &item.PublicationsCount,
+		&item.RetryCount, &item.MaxRetries, &item.LastError,
+		&item.StartedAt, &item.CompletedAt,
+		&item.CreatedAt, &item.UpdatedAt,
+	)
+
+	if err == pgx.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, fmt.Errorf("failed to get next queue item: %w", err)
+	}
+
+	item.CurrentPhase = CrawlPhase(phase)
+	return &item, nil
+}
+
+// AddToQueue adds a university to the crawl queue
+func (r *PostgresRepository) AddToQueue(ctx context.Context, universityID uuid.UUID, priority int, initiatedBy string) (*CrawlQueueItem, error) {
+	// Get next queue position
+	var nextPosition int
+	err := r.pool.QueryRow(ctx, `SELECT COALESCE(MAX(queue_position), 0) + 1 FROM crawl_queue WHERE queue_position IS NOT NULL`).Scan(&nextPosition)
+	if err != nil {
+		return nil, fmt.Errorf("failed to get next queue position: %w", err)
+	}
+
+	query := `
+		INSERT INTO crawl_queue (university_id, queue_position, priority, initiated_by)
+		VALUES ($1, $2, $3, $4)
+		ON CONFLICT (university_id) DO UPDATE SET
+			queue_position = EXCLUDED.queue_position,
+			priority = EXCLUDED.priority,
+			current_phase = 'pending',
+			retry_count = 0,
+			last_error = NULL,
+			updated_at = NOW()
+		RETURNING id, created_at, updated_at
+	`
+
+	item := &CrawlQueueItem{
+		UniversityID:  universityID,
+		QueuePosition: &nextPosition,
+		Priority:      priority,
+		CurrentPhase:  PhasePending,
+		MaxRetries:    3,
+	}
+
+	err = r.pool.QueryRow(ctx, query, universityID, nextPosition, priority, initiatedBy).Scan(
+		&item.ID, &item.CreatedAt, &item.UpdatedAt,
+	)
+	if err != nil {
+		return nil, fmt.Errorf("failed to add to queue: %w", err)
+	}
+
+	// Get university name
+	r.pool.QueryRow(ctx, `SELECT name, short_name FROM universities WHERE id = $1`, universityID).Scan(
+		&item.UniversityName, &item.UniversityShort,
+	)
+
+	return item, nil
+}
+
+// RemoveFromQueue removes a university from the queue
+func (r *PostgresRepository) RemoveFromQueue(ctx context.Context, universityID uuid.UUID) error {
+	_, err := r.pool.Exec(ctx, `DELETE FROM crawl_queue WHERE university_id = $1`, universityID)
+	return err
+}
+
+// UpdateQueueItem updates a queue item
+func (r *PostgresRepository) UpdateQueueItem(ctx context.Context, item *CrawlQueueItem) error {
+	query := `
+		UPDATE crawl_queue SET
+			queue_position = $2,
+			priority = $3,
+			current_phase = $4,
+			discovery_completed = $5,
+			discovery_completed_at = $6,
+			professors_completed = $7,
+			professors_completed_at = $8,
+			all_staff_completed = $9,
+			all_staff_completed_at = $10,
+			publications_completed = $11,
+			publications_completed_at = $12,
+			discovery_count = $13,
+			professors_count = $14,
+			staff_count = $15,
+			publications_count = $16,
+			retry_count = $17,
+			last_error = $18,
+			started_at = $19,
+			completed_at = $20,
+			updated_at = NOW()
+		WHERE university_id = $1
+	`
+
+	_, err := r.pool.Exec(ctx, query,
+		item.UniversityID,
+		item.QueuePosition, item.Priority, string(item.CurrentPhase),
+		item.DiscoveryCompleted, item.DiscoveryCompletedAt,
+		item.ProfessorsCompleted, item.ProfessorsCompletedAt,
+		item.AllStaffCompleted, item.AllStaffCompletedAt,
+		item.PublicationsCompleted, item.PublicationsCompletedAt,
+		item.DiscoveryCount, item.ProfessorsCount, item.StaffCount, item.PublicationsCount,
+		item.RetryCount, item.LastError,
+		item.StartedAt, item.CompletedAt,
+	)
+	return err
+}
+
+// PauseQueueItem pauses a crawl
+func (r *PostgresRepository) PauseQueueItem(ctx context.Context, universityID uuid.UUID) error {
+	_, err := r.pool.Exec(ctx, `UPDATE crawl_queue SET current_phase = 'paused', updated_at = NOW() WHERE university_id = $1`, universityID)
+	return err
+}
+
+// ResumeQueueItem resumes a paused crawl
+func (r *PostgresRepository) ResumeQueueItem(ctx context.Context, universityID uuid.UUID) error {
+	// Determine what phase to resume from
+	query := `
+		UPDATE crawl_queue SET
+			current_phase = CASE
+				WHEN NOT discovery_completed THEN 'discovery'
+				WHEN NOT professors_completed THEN 'professors'
+				WHEN NOT all_staff_completed THEN 'all_staff'
+				WHEN NOT publications_completed THEN 'publications'
+				ELSE 'pending'
+			END,
+			updated_at = NOW()
+		WHERE university_id = $1 AND current_phase = 'paused'
+	`
+	_, err := r.pool.Exec(ctx, query, universityID)
+	return err
+}
+
+// ============================================================================
+// PHASE UPDATES
+// ============================================================================
+
+// CompletePhase marks a phase as completed
+func (r *PostgresRepository) CompletePhase(ctx context.Context, universityID uuid.UUID, phase CrawlPhase, count int) error {
+	now := time.Now()
+	var query string
+
+	switch phase {
+	case PhaseDiscovery:
+		query = `UPDATE crawl_queue SET discovery_completed = true, discovery_completed_at = $2, discovery_count = $3, updated_at = NOW() WHERE university_id = $1`
+	case PhaseProfessors:
+		query = `UPDATE crawl_queue SET professors_completed = true, professors_completed_at = $2, professors_count = $3, updated_at = NOW() WHERE university_id = $1`
+	case PhaseAllStaff:
+		query = `UPDATE crawl_queue SET all_staff_completed = true, all_staff_completed_at = $2, staff_count = $3, updated_at = NOW() WHERE university_id = $1`
+	case PhasePublications:
+		query = `UPDATE crawl_queue SET publications_completed = true, publications_completed_at = $2, publications_count = $3, updated_at = NOW() WHERE university_id = $1`
+	default:
+		return fmt.Errorf("unknown phase: %s", phase)
+	}
+
+	_, err := r.pool.Exec(ctx, query, universityID, now, count)
+	return err
+}
+
+// FailPhase records a phase failure
+func (r *PostgresRepository) FailPhase(ctx context.Context, universityID uuid.UUID, phase CrawlPhase, errMsg string) error {
+	query := `
+		UPDATE crawl_queue SET
+			retry_count = retry_count + 1,
+			last_error = $2,
+			current_phase = CASE
+				WHEN retry_count + 1 >= max_retries THEN 'failed'
+				ELSE current_phase
+			END,
+			updated_at = NOW()
+		WHERE university_id = $1
+	`
+	_, err := r.pool.Exec(ctx, query, universityID, errMsg)
+	return err
+}
+
+// ============================================================================
+// STATS
+// ============================================================================
+
+// GetCompletedTodayCount returns the number of universities completed today
+func (r *PostgresRepository) GetCompletedTodayCount(ctx context.Context) (int, error) {
+	var count int
+	err := r.pool.QueryRow(ctx, `
+		SELECT COUNT(*) FROM crawl_queue
+		WHERE current_phase = 'completed'
+		AND completed_at >= CURRENT_DATE
+	`).Scan(&count)
+	return count, err
+}
+
+// GetTotalProcessedCount returns the total number of processed universities
+func (r *PostgresRepository) GetTotalProcessedCount(ctx context.Context) (int, error) {
+	var count int
+	err := r.pool.QueryRow(ctx, `SELECT COUNT(*) FROM crawl_queue WHERE current_phase = 'completed'`).Scan(&count)
+	return count, err
+}
--- a/edu-search-service/internal/pipeline/pipeline.go
+++ b/edu-search-service/internal/pipeline/pipeline.go
@@ -0,0 +1,301 @@
+package pipeline
+
+import (
+	"context"
+	"log"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/breakpilot/edu-search-service/internal/crawler"
+	"github.com/breakpilot/edu-search-service/internal/extractor"
+	"github.com/breakpilot/edu-search-service/internal/indexer"
+	"github.com/breakpilot/edu-search-service/internal/tagger"
+)
+
+// Pipeline orchestrates crawling, extraction, tagging, and indexing
+type Pipeline struct {
+	crawler     *crawler.Crawler
+	tagger      *tagger.Tagger
+	indexClient *indexer.Client
+	maxPages    int
+	workers     int
+}
+
+// Stats tracks pipeline execution statistics
+type Stats struct {
+	StartTime      time.Time
+	EndTime        time.Time
+	URLsProcessed  int
+	URLsSuccessful int
+	URLsFailed     int
+	URLsSkipped    int
+	DocumentsIndexed int
+}
+
+// NewPipeline creates a new crawl pipeline
+func NewPipeline(
+	crawlerInstance *crawler.Crawler,
+	taggerInstance *tagger.Tagger,
+	indexClient *indexer.Client,
+	maxPages int,
+) *Pipeline {
+	return &Pipeline{
+		crawler:     crawlerInstance,
+		tagger:      taggerInstance,
+		indexClient: indexClient,
+		maxPages:    maxPages,
+		workers:     5, // concurrent workers
+	}
+}
+
+// Run executes the crawl pipeline
+func (p *Pipeline) Run(ctx context.Context, seedsDir string) (*Stats, error) {
+	stats := &Stats{
+		StartTime: time.Now(),
+	}
+
+	// Load seed URLs
+	seeds, err := p.crawler.LoadSeeds(seedsDir)
+	if err != nil {
+		return nil, err
+	}
+
+	log.Printf("Pipeline starting with %d seeds, max %d pages", len(seeds), p.maxPages)
+
+	// Create URL queue
+	urlQueue := make(chan string, len(seeds)*10)
+	visited := &sync.Map{}
+
+	// Add seeds to queue
+	for _, seed := range seeds {
+		normalized := crawler.NormalizeURL(seed)
+		if _, loaded := visited.LoadOrStore(normalized, true); !loaded {
+			urlQueue <- seed
+		}
+	}
+
+	// Results channel
+	results := make(chan *processResult, p.workers*2)
+	var wg sync.WaitGroup
+
+	// Start workers
+	for i := 0; i < p.workers; i++ {
+		wg.Add(1)
+		go p.worker(ctx, i, urlQueue, results, visited, &wg)
+	}
+
+	// Close results when all workers done
+	go func() {
+		wg.Wait()
+		close(results)
+	}()
+
+	// Process results and collect stats
+	var documents []indexer.Document
+	processed := 0
+
+	for result := range results {
+		stats.URLsProcessed++
+
+		if result.err != nil {
+			stats.URLsFailed++
+			continue
+		}
+
+		if result.skipped {
+			stats.URLsSkipped++
+			continue
+		}
+
+		if result.document != nil {
+			documents = append(documents, *result.document)
+			stats.URLsSuccessful++
+
+			// Bulk index every 50 documents
+			if len(documents) >= 50 {
+				if err := p.indexClient.BulkIndex(ctx, documents); err != nil {
+					log.Printf("Bulk index error: %v", err)
+				} else {
+					stats.DocumentsIndexed += len(documents)
+				}
+				documents = nil
+			}
+		}
+
+		processed++
+		if processed >= p.maxPages {
+			log.Printf("Reached max pages limit (%d)", p.maxPages)
+			close(urlQueue)
+			break
+		}
+	}
+
+	// Index remaining documents
+	if len(documents) > 0 {
+		if err := p.indexClient.BulkIndex(ctx, documents); err != nil {
+			log.Printf("Final bulk index error: %v", err)
+		} else {
+			stats.DocumentsIndexed += len(documents)
+		}
+	}
+
+	stats.EndTime = time.Now()
+	log.Printf("Pipeline completed: %d processed, %d indexed, %d failed, %d skipped in %v",
+		stats.URLsProcessed, stats.DocumentsIndexed, stats.URLsFailed, stats.URLsSkipped,
+		stats.EndTime.Sub(stats.StartTime))
+
+	return stats, nil
+}
+
+type processResult struct {
+	url      string
+	document *indexer.Document
+	err      error
+	skipped  bool
+}
+
+func (p *Pipeline) worker(
+	ctx context.Context,
+	id int,
+	urlQueue chan string,
+	results chan<- *processResult,
+	visited *sync.Map,
+	wg *sync.WaitGroup,
+) {
+	defer wg.Done()
+
+	for url := range urlQueue {
+		select {
+		case <-ctx.Done():
+			return
+		default:
+			result := p.processURL(ctx, url, urlQueue, visited)
+			results <- result
+		}
+	}
+}
+
+func (p *Pipeline) processURL(
+	ctx context.Context,
+	url string,
+	urlQueue chan<- string,
+	visited *sync.Map,
+) *processResult {
+	result := &processResult{url: url}
+
+	// Fetch URL
+	fetchResult, err := p.crawler.Fetch(ctx, url)
+	if err != nil {
+		result.err = err
+		return result
+	}
+
+	// Check content type
+	contentType := strings.ToLower(fetchResult.ContentType)
+	if !strings.Contains(contentType, "text/html") && !strings.Contains(contentType, "application/pdf") {
+		result.skipped = true
+		return result
+	}
+
+	// Extract content
+	var extracted *extractor.ExtractedContent
+	if strings.Contains(contentType, "text/html") {
+		extracted, err = extractor.ExtractHTML(fetchResult.Body)
+	} else if strings.Contains(contentType, "application/pdf") {
+		extracted, err = extractor.ExtractPDF(fetchResult.Body)
+	}
+
+	if err != nil {
+		result.err = err
+		return result
+	}
+
+	// Skip if too little content
+	if extracted.ContentLength < 100 {
+		result.skipped = true
+		return result
+	}
+
+	// Tag content
+	features := tagger.ContentFeatures{
+		AdDensity:     extracted.Features.AdDensity,
+		LinkDensity:   extracted.Features.LinkDensity,
+		ContentLength: extracted.ContentLength,
+	}
+	tags := p.tagger.Tag(fetchResult.CanonicalURL, extracted.Title, extracted.ContentText, features)
+
+	// Create document
+	doc := &indexer.Document{
+		DocID:        crawler.GenerateDocID(),
+		URL:          fetchResult.CanonicalURL,
+		Domain:       crawler.ExtractDomain(fetchResult.CanonicalURL),
+		Title:        extracted.Title,
+		ContentText:  extracted.ContentText,
+		SnippetText:  extracted.SnippetText,
+		ContentHash:  fetchResult.ContentHash,
+		DocType:      tags.DocType,
+		Subjects:     tags.Subjects,
+		SchoolLevel:  tags.SchoolLevel,
+		State:        tags.State,
+		Language:     extracted.Language,
+		TrustScore:   tags.TrustScore,
+		QualityScore: calculateQualityScore(extracted, tags),
+		FetchedAt:    fetchResult.FetchTime,
+		UpdatedAt:    time.Now(),
+	}
+
+	result.document = doc
+
+	// Extract and queue new links (limited to same domain for now)
+	docDomain := crawler.ExtractDomain(url)
+	for _, link := range extracted.Links {
+		linkDomain := crawler.ExtractDomain(link)
+		if linkDomain == docDomain {
+			normalized := crawler.NormalizeURL(link)
+			if _, loaded := visited.LoadOrStore(normalized, true); !loaded {
+				select {
+				case urlQueue <- link:
+				default:
+					// Queue full, skip
+				}
+			}
+		}
+	}
+
+	return result
+}
+
+func calculateQualityScore(extracted *extractor.ExtractedContent, tags tagger.TagResult) float64 {
+	score := 0.5 // base
+
+	// Content length bonus
+	if extracted.ContentLength > 1000 {
+		score += 0.1
+	}
+	if extracted.ContentLength > 5000 {
+		score += 0.1
+	}
+
+	// Has headings
+	if len(extracted.Headings) > 0 {
+		score += 0.1
+	}
+
+	// Low ad density
+	if extracted.Features.AdDensity < 0.1 {
+		score += 0.1
+	}
+
+	// Good text/HTML ratio
+	if extracted.Features.TextToHTMLRatio > 0.2 {
+		score += 0.1
+	}
+
+	// Clamp
+	if score > 1 {
+		score = 1
+	}
+
+	return score
+}
--- a/edu-search-service/internal/policy/audit.go
+++ b/edu-search-service/internal/policy/audit.go
@@ -0,0 +1,255 @@
+package policy
+
+import (
+	"context"
+	"encoding/json"
+
+	"github.com/google/uuid"
+)
+
+// Auditor provides audit logging functionality for the policy system.
+type Auditor struct {
+	store *Store
+}
+
+// NewAuditor creates a new Auditor instance.
+func NewAuditor(store *Store) *Auditor {
+	return &Auditor{store: store}
+}
+
+// LogChange logs a policy change to the audit trail.
+func (a *Auditor) LogChange(ctx context.Context, action AuditAction, entityType AuditEntityType, entityID *uuid.UUID, oldValue, newValue interface{}, userEmail, ipAddress, userAgent *string) error {
+	entry := &PolicyAuditLog{
+		Action:     action,
+		EntityType: entityType,
+		EntityID:   entityID,
+		UserEmail:  userEmail,
+		IPAddress:  ipAddress,
+		UserAgent:  userAgent,
+	}
+
+	if oldValue != nil {
+		entry.OldValue = toJSON(oldValue)
+	}
+	if newValue != nil {
+		entry.NewValue = toJSON(newValue)
+	}
+
+	return a.store.CreateAuditLog(ctx, entry)
+}
+
+// LogBlocked logs a blocked URL to the blocked content log.
+func (a *Auditor) LogBlocked(ctx context.Context, url, domain string, reason BlockReason, ruleID *uuid.UUID, details map[string]interface{}) error {
+	entry := &BlockedContentLog{
+		URL:           url,
+		Domain:        domain,
+		BlockReason:   reason,
+		MatchedRuleID: ruleID,
+	}
+
+	if details != nil {
+		entry.Details = toJSON(details)
+	}
+
+	return a.store.CreateBlockedContentLog(ctx, entry)
+}
+
+// =============================================================================
+// CONVENIENCE METHODS
+// =============================================================================
+
+// LogPolicyCreated logs a policy creation event.
+func (a *Auditor) LogPolicyCreated(ctx context.Context, policy *SourcePolicy, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionCreate, AuditEntitySourcePolicy, &policy.ID, nil, policy, userEmail, nil, nil)
+}
+
+// LogPolicyUpdated logs a policy update event.
+func (a *Auditor) LogPolicyUpdated(ctx context.Context, oldPolicy, newPolicy *SourcePolicy, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionUpdate, AuditEntitySourcePolicy, &newPolicy.ID, oldPolicy, newPolicy, userEmail, nil, nil)
+}
+
+// LogPolicyDeleted logs a policy deletion event.
+func (a *Auditor) LogPolicyDeleted(ctx context.Context, policy *SourcePolicy, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionDelete, AuditEntitySourcePolicy, &policy.ID, policy, nil, userEmail, nil, nil)
+}
+
+// LogPolicyActivated logs a policy activation event.
+func (a *Auditor) LogPolicyActivated(ctx context.Context, policy *SourcePolicy, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionActivate, AuditEntitySourcePolicy, &policy.ID, nil, policy, userEmail, nil, nil)
+}
+
+// LogPolicyDeactivated logs a policy deactivation event.
+func (a *Auditor) LogPolicyDeactivated(ctx context.Context, policy *SourcePolicy, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionDeactivate, AuditEntitySourcePolicy, &policy.ID, policy, nil, userEmail, nil, nil)
+}
+
+// LogSourceCreated logs a source creation event.
+func (a *Auditor) LogSourceCreated(ctx context.Context, source *AllowedSource, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionCreate, AuditEntityAllowedSource, &source.ID, nil, source, userEmail, nil, nil)
+}
+
+// LogSourceUpdated logs a source update event.
+func (a *Auditor) LogSourceUpdated(ctx context.Context, oldSource, newSource *AllowedSource, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionUpdate, AuditEntityAllowedSource, &newSource.ID, oldSource, newSource, userEmail, nil, nil)
+}
+
+// LogSourceDeleted logs a source deletion event.
+func (a *Auditor) LogSourceDeleted(ctx context.Context, source *AllowedSource, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionDelete, AuditEntityAllowedSource, &source.ID, source, nil, userEmail, nil, nil)
+}
+
+// LogOperationUpdated logs an operation permission update event.
+func (a *Auditor) LogOperationUpdated(ctx context.Context, oldOp, newOp *OperationPermission, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionUpdate, AuditEntityOperationPermission, &newOp.ID, oldOp, newOp, userEmail, nil, nil)
+}
+
+// LogPIIRuleCreated logs a PII rule creation event.
+func (a *Auditor) LogPIIRuleCreated(ctx context.Context, rule *PIIRule, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionCreate, AuditEntityPIIRule, &rule.ID, nil, rule, userEmail, nil, nil)
+}
+
+// LogPIIRuleUpdated logs a PII rule update event.
+func (a *Auditor) LogPIIRuleUpdated(ctx context.Context, oldRule, newRule *PIIRule, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionUpdate, AuditEntityPIIRule, &newRule.ID, oldRule, newRule, userEmail, nil, nil)
+}
+
+// LogPIIRuleDeleted logs a PII rule deletion event.
+func (a *Auditor) LogPIIRuleDeleted(ctx context.Context, rule *PIIRule, userEmail *string) error {
+	return a.LogChange(ctx, AuditActionDelete, AuditEntityPIIRule, &rule.ID, rule, nil, userEmail, nil, nil)
+}
+
+// LogContentBlocked logs a blocked content event with details.
+func (a *Auditor) LogContentBlocked(ctx context.Context, url, domain string, reason BlockReason, matchedPatterns []string, ruleID *uuid.UUID) error {
+	details := map[string]interface{}{
+		"matched_patterns": matchedPatterns,
+	}
+	return a.LogBlocked(ctx, url, domain, reason, ruleID, details)
+}
+
+// LogPIIBlocked logs content blocked due to PII detection.
+func (a *Auditor) LogPIIBlocked(ctx context.Context, url, domain string, matches []PIIMatch) error {
+	matchDetails := make([]map[string]interface{}, len(matches))
+	var ruleID *uuid.UUID
+
+	for i, m := range matches {
+		matchDetails[i] = map[string]interface{}{
+			"rule_name": m.RuleName,
+			"severity":  m.Severity,
+			"match":     maskPII(m.Match), // Mask the actual PII in logs
+		}
+		if ruleID == nil {
+			ruleID = &m.RuleID
+		}
+	}
+
+	details := map[string]interface{}{
+		"pii_matches": matchDetails,
+		"match_count": len(matches),
+	}
+
+	return a.LogBlocked(ctx, url, domain, BlockReasonPIIDetected, ruleID, details)
+}
+
+// =============================================================================
+// HELPERS
+// =============================================================================
+
+// toJSON converts a value to JSON.
+func toJSON(v interface{}) json.RawMessage {
+	data, err := json.Marshal(v)
+	if err != nil {
+		return nil
+	}
+	return data
+}
+
+// maskPII masks PII data for safe logging.
+func maskPII(pii string) string {
+	if len(pii) <= 4 {
+		return "****"
+	}
+	// Show first 2 and last 2 characters
+	return pii[:2] + "****" + pii[len(pii)-2:]
+}
+
+// =============================================================================
+// AUDIT REPORT GENERATION
+// =============================================================================
+
+// AuditReport represents an audit report for compliance.
+type AuditReport struct {
+	GeneratedAt    string              `json:"generated_at"`
+	PeriodStart    string              `json:"period_start"`
+	PeriodEnd      string              `json:"period_end"`
+	Summary        AuditReportSummary  `json:"summary"`
+	PolicyChanges  []PolicyAuditLog    `json:"policy_changes"`
+	BlockedContent []BlockedContentLog `json:"blocked_content"`
+	Stats          *PolicyStats        `json:"stats"`
+}
+
+// AuditReportSummary contains summary statistics for the audit report.
+type AuditReportSummary struct {
+	TotalPolicyChanges int            `json:"total_policy_changes"`
+	TotalBlocked       int            `json:"total_blocked"`
+	ChangesByAction    map[string]int `json:"changes_by_action"`
+	BlocksByReason     map[string]int `json:"blocks_by_reason"`
+}
+
+// GenerateAuditReport generates a compliance audit report.
+func (a *Auditor) GenerateAuditReport(ctx context.Context, filter *AuditLogFilter, blockedFilter *BlockedContentFilter) (*AuditReport, error) {
+	// Get audit logs
+	auditLogs, _, err := a.store.ListAuditLogs(ctx, filter)
+	if err != nil {
+		return nil, err
+	}
+
+	// Get blocked content
+	blockedLogs, _, err := a.store.ListBlockedContent(ctx, blockedFilter)
+	if err != nil {
+		return nil, err
+	}
+
+	// Get stats
+	stats, err := a.store.GetStats(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	// Build summary
+	summary := AuditReportSummary{
+		TotalPolicyChanges: len(auditLogs),
+		TotalBlocked:       len(blockedLogs),
+		ChangesByAction:    make(map[string]int),
+		BlocksByReason:     make(map[string]int),
+	}
+
+	for _, log := range auditLogs {
+		summary.ChangesByAction[string(log.Action)]++
+	}
+
+	for _, log := range blockedLogs {
+		summary.BlocksByReason[string(log.BlockReason)]++
+	}
+
+	// Build report
+	periodStart := ""
+	periodEnd := ""
+	if filter.FromDate != nil {
+		periodStart = filter.FromDate.Format("2006-01-02")
+	}
+	if filter.ToDate != nil {
+		periodEnd = filter.ToDate.Format("2006-01-02")
+	}
+
+	report := &AuditReport{
+		GeneratedAt:    uuid.New().String()[:19], // Timestamp placeholder
+		PeriodStart:    periodStart,
+		PeriodEnd:      periodEnd,
+		Summary:        summary,
+		PolicyChanges:  auditLogs,
+		BlockedContent: blockedLogs,
+		Stats:          stats,
+	}
+
+	return report, nil
+}
--- a/edu-search-service/internal/policy/enforcer.go
+++ b/edu-search-service/internal/policy/enforcer.go
@@ -0,0 +1,281 @@
+package policy
+
+import (
+	"context"
+	"net/url"
+	"strings"
+
+	"github.com/google/uuid"
+)
+
+// Enforcer provides policy enforcement for the crawler and pipeline.
+type Enforcer struct {
+	store       *Store
+	piiDetector *PIIDetector
+	auditor     *Auditor
+}
+
+// NewEnforcer creates a new Enforcer instance.
+func NewEnforcer(store *Store) *Enforcer {
+	return &Enforcer{
+		store:       store,
+		piiDetector: NewPIIDetector(store),
+		auditor:     NewAuditor(store),
+	}
+}
+
+// =============================================================================
+// SOURCE CHECKING
+// =============================================================================
+
+// CheckSource verifies if a URL is allowed based on the whitelist.
+// Returns the AllowedSource if found, nil if not whitelisted.
+func (e *Enforcer) CheckSource(ctx context.Context, rawURL string, bundesland *Bundesland) (*AllowedSource, error) {
+	domain, err := extractDomain(rawURL)
+	if err != nil {
+		return nil, err
+	}
+
+	source, err := e.store.GetSourceByDomain(ctx, domain, bundesland)
+	if err != nil {
+		return nil, err
+	}
+
+	return source, nil
+}
+
+// CheckOperation verifies if a specific operation is allowed for a source.
+func (e *Enforcer) CheckOperation(ctx context.Context, source *AllowedSource, operation Operation) (*OperationPermission, error) {
+	for _, op := range source.Operations {
+		if op.Operation == operation {
+			return &op, nil
+		}
+	}
+
+	// If not found in loaded operations, query directly
+	ops, err := e.store.GetOperationsBySourceID(ctx, source.ID)
+	if err != nil {
+		return nil, err
+	}
+
+	for _, op := range ops {
+		if op.Operation == operation {
+			return &op, nil
+		}
+	}
+
+	return nil, nil
+}
+
+// CheckCompliance performs a full compliance check for a URL and operation.
+func (e *Enforcer) CheckCompliance(ctx context.Context, req *CheckComplianceRequest) (*CheckComplianceResponse, error) {
+	response := &CheckComplianceResponse{
+		IsAllowed:        false,
+		RequiresCitation: false,
+	}
+
+	// Check if source is whitelisted
+	source, err := e.CheckSource(ctx, req.URL, req.Bundesland)
+	if err != nil {
+		return nil, err
+	}
+
+	if source == nil {
+		reason := BlockReasonNotWhitelisted
+		response.BlockReason = &reason
+		return response, nil
+	}
+
+	response.Source = source
+	response.License = &source.License
+	response.CitationTemplate = source.CitationTemplate
+
+	// Check operation permission
+	opPerm, err := e.CheckOperation(ctx, source, req.Operation)
+	if err != nil {
+		return nil, err
+	}
+
+	if opPerm == nil || !opPerm.IsAllowed {
+		var reason BlockReason
+		if req.Operation == OperationTraining {
+			reason = BlockReasonTrainingForbidden
+		} else {
+			reason = BlockReasonLicenseViolation
+		}
+		response.BlockReason = &reason
+		return response, nil
+	}
+
+	response.IsAllowed = true
+	response.RequiresCitation = opPerm.RequiresCitation
+
+	return response, nil
+}
+
+// =============================================================================
+// PII CHECKING
+// =============================================================================
+
+// DetectPII scans text for PII patterns and returns matches.
+func (e *Enforcer) DetectPII(ctx context.Context, text string) (*PIITestResponse, error) {
+	return e.piiDetector.Detect(ctx, text)
+}
+
+// ShouldBlockForPII determines if content should be blocked based on PII matches.
+func (e *Enforcer) ShouldBlockForPII(response *PIITestResponse) bool {
+	if response == nil {
+		return false
+	}
+	return response.ShouldBlock
+}
+
+// =============================================================================
+// LOGGING
+// =============================================================================
+
+// LogBlocked logs a blocked URL to the blocked content log.
+func (e *Enforcer) LogBlocked(ctx context.Context, rawURL string, reason BlockReason, ruleID *uuid.UUID, details map[string]interface{}) error {
+	domain, _ := extractDomain(rawURL)
+	return e.auditor.LogBlocked(ctx, rawURL, domain, reason, ruleID, details)
+}
+
+// LogChange logs a policy change to the audit log.
+func (e *Enforcer) LogChange(ctx context.Context, action AuditAction, entityType AuditEntityType, entityID *uuid.UUID, oldValue, newValue interface{}, userEmail *string) error {
+	return e.auditor.LogChange(ctx, action, entityType, entityID, oldValue, newValue, userEmail, nil, nil)
+}
+
+// =============================================================================
+// BATCH OPERATIONS
+// =============================================================================
+
+// FilterURLs filters a list of URLs, returning only whitelisted ones.
+func (e *Enforcer) FilterURLs(ctx context.Context, urls []string, bundesland *Bundesland, operation Operation) ([]FilteredURL, error) {
+	results := make([]FilteredURL, 0, len(urls))
+
+	for _, u := range urls {
+		result := FilteredURL{
+			URL:       u,
+			IsAllowed: false,
+		}
+
+		source, err := e.CheckSource(ctx, u, bundesland)
+		if err != nil {
+			result.Error = err.Error()
+			results = append(results, result)
+			continue
+		}
+
+		if source == nil {
+			result.BlockReason = BlockReasonNotWhitelisted
+			results = append(results, result)
+			continue
+		}
+
+		opPerm, err := e.CheckOperation(ctx, source, operation)
+		if err != nil {
+			result.Error = err.Error()
+			results = append(results, result)
+			continue
+		}
+
+		if opPerm == nil || !opPerm.IsAllowed {
+			if operation == OperationTraining {
+				result.BlockReason = BlockReasonTrainingForbidden
+			} else {
+				result.BlockReason = BlockReasonLicenseViolation
+			}
+			results = append(results, result)
+			continue
+		}
+
+		result.IsAllowed = true
+		result.Source = source
+		result.RequiresCitation = opPerm.RequiresCitation
+		results = append(results, result)
+	}
+
+	return results, nil
+}
+
+// FilteredURL represents the result of filtering a single URL.
+type FilteredURL struct {
+	URL              string         `json:"url"`
+	IsAllowed        bool           `json:"is_allowed"`
+	Source           *AllowedSource `json:"source,omitempty"`
+	BlockReason      BlockReason    `json:"block_reason,omitempty"`
+	RequiresCitation bool           `json:"requires_citation"`
+	Error            string         `json:"error,omitempty"`
+}
+
+// =============================================================================
+// HELPERS
+// =============================================================================
+
+// extractDomain extracts the domain from a URL.
+func extractDomain(rawURL string) (string, error) {
+	// Handle URLs without scheme
+	if !strings.Contains(rawURL, "://") {
+		rawURL = "https://" + rawURL
+	}
+
+	parsed, err := url.Parse(rawURL)
+	if err != nil {
+		return "", err
+	}
+
+	host := parsed.Hostname()
+
+	// Remove www. prefix
+	host = strings.TrimPrefix(host, "www.")
+
+	return host, nil
+}
+
+// IsTrainingAllowed checks if training is allowed for any source (should always be false).
+func (e *Enforcer) IsTrainingAllowed(ctx context.Context) (bool, error) {
+	// Training should NEVER be allowed - this is a safeguard
+	matrix, err := e.store.GetOperationsMatrix(ctx)
+	if err != nil {
+		return false, err
+	}
+
+	for _, source := range matrix {
+		for _, op := range source.Operations {
+			if op.Operation == OperationTraining && op.IsAllowed {
+				// This should never happen - log a warning
+				return true, nil
+			}
+		}
+	}
+
+	return false, nil
+}
+
+// GetSourceByURL is a convenience method to get a source by URL.
+func (e *Enforcer) GetSourceByURL(ctx context.Context, rawURL string, bundesland *Bundesland) (*AllowedSource, error) {
+	return e.CheckSource(ctx, rawURL, bundesland)
+}
+
+// GetCitationForURL generates a citation for a URL if required.
+func (e *Enforcer) GetCitationForURL(ctx context.Context, rawURL string, bundesland *Bundesland, title string, date string) (string, error) {
+	source, err := e.CheckSource(ctx, rawURL, bundesland)
+	if err != nil || source == nil {
+		return "", err
+	}
+
+	if source.CitationTemplate == nil || *source.CitationTemplate == "" {
+		// Default citation format
+		return "Quelle: " + source.Name + ", " + title + ", " + date, nil
+	}
+
+	// Replace placeholders in template
+	citation := *source.CitationTemplate
+	citation = strings.ReplaceAll(citation, "{title}", title)
+	citation = strings.ReplaceAll(citation, "{date}", date)
+	citation = strings.ReplaceAll(citation, "{url}", rawURL)
+	citation = strings.ReplaceAll(citation, "{domain}", source.Domain)
+	citation = strings.ReplaceAll(citation, "{source}", source.Name)
+
+	return citation, nil
+}
--- a/edu-search-service/internal/policy/loader.go
+++ b/edu-search-service/internal/policy/loader.go
@@ -0,0 +1,255 @@
+package policy
+
+import (
+	"context"
+	"fmt"
+	"os"
+
+	"gopkg.in/yaml.v3"
+)
+
+// Loader handles loading policy configuration from YAML files.
+type Loader struct {
+	store *Store
+}
+
+// NewLoader creates a new Loader instance.
+func NewLoader(store *Store) *Loader {
+	return &Loader{store: store}
+}
+
+// LoadFromFile loads policy configuration from a YAML file.
+func (l *Loader) LoadFromFile(ctx context.Context, path string) error {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return fmt.Errorf("failed to read YAML file: %w", err)
+	}
+
+	config, err := ParseYAML(data)
+	if err != nil {
+		return fmt.Errorf("failed to parse YAML: %w", err)
+	}
+
+	return l.store.LoadFromYAML(ctx, config)
+}
+
+// ParseYAML parses YAML configuration data.
+func ParseYAML(data []byte) (*BundeslaenderConfig, error) {
+	// First, parse as a generic map to handle the inline Bundeslaender
+	var rawConfig map[string]interface{}
+	if err := yaml.Unmarshal(data, &rawConfig); err != nil {
+		return nil, fmt.Errorf("failed to parse YAML: %w", err)
+	}
+
+	config := &BundeslaenderConfig{
+		Bundeslaender: make(map[string]PolicyConfig),
+	}
+
+	// Parse federal
+	if federal, ok := rawConfig["federal"]; ok {
+		if federalMap, ok := federal.(map[string]interface{}); ok {
+			config.Federal = parsePolicyConfig(federalMap)
+		}
+	}
+
+	// Parse default_operations
+	if ops, ok := rawConfig["default_operations"]; ok {
+		if opsMap, ok := ops.(map[string]interface{}); ok {
+			config.DefaultOperations = parseOperationsConfig(opsMap)
+		}
+	}
+
+	// Parse pii_rules
+	if rules, ok := rawConfig["pii_rules"]; ok {
+		if rulesSlice, ok := rules.([]interface{}); ok {
+			for _, rule := range rulesSlice {
+				if ruleMap, ok := rule.(map[string]interface{}); ok {
+					config.PIIRules = append(config.PIIRules, parsePIIRuleConfig(ruleMap))
+				}
+			}
+		}
+	}
+
+	// Parse Bundeslaender (2-letter codes)
+	bundeslaender := []string{"BW", "BY", "BE", "BB", "HB", "HH", "HE", "MV", "NI", "NW", "RP", "SL", "SN", "ST", "SH", "TH"}
+	for _, bl := range bundeslaender {
+		if blConfig, ok := rawConfig[bl]; ok {
+			if blMap, ok := blConfig.(map[string]interface{}); ok {
+				config.Bundeslaender[bl] = parsePolicyConfig(blMap)
+			}
+		}
+	}
+
+	return config, nil
+}
+
+func parsePolicyConfig(m map[string]interface{}) PolicyConfig {
+	pc := PolicyConfig{}
+
+	if name, ok := m["name"].(string); ok {
+		pc.Name = name
+	}
+
+	if sources, ok := m["sources"].([]interface{}); ok {
+		for _, src := range sources {
+			if srcMap, ok := src.(map[string]interface{}); ok {
+				pc.Sources = append(pc.Sources, parseSourceConfig(srcMap))
+			}
+		}
+	}
+
+	return pc
+}
+
+func parseSourceConfig(m map[string]interface{}) SourceConfig {
+	sc := SourceConfig{
+		TrustBoost: 0.5, // Default
+	}
+
+	if domain, ok := m["domain"].(string); ok {
+		sc.Domain = domain
+	}
+	if name, ok := m["name"].(string); ok {
+		sc.Name = name
+	}
+	if license, ok := m["license"].(string); ok {
+		sc.License = license
+	}
+	if legalBasis, ok := m["legal_basis"].(string); ok {
+		sc.LegalBasis = legalBasis
+	}
+	if citation, ok := m["citation_template"].(string); ok {
+		sc.CitationTemplate = citation
+	}
+	if trustBoost, ok := m["trust_boost"].(float64); ok {
+		sc.TrustBoost = trustBoost
+	}
+
+	return sc
+}
+
+func parseOperationsConfig(m map[string]interface{}) OperationsConfig {
+	oc := OperationsConfig{}
+
+	if lookup, ok := m["lookup"].(map[string]interface{}); ok {
+		oc.Lookup = parseOperationConfig(lookup)
+	}
+	if rag, ok := m["rag"].(map[string]interface{}); ok {
+		oc.RAG = parseOperationConfig(rag)
+	}
+	if training, ok := m["training"].(map[string]interface{}); ok {
+		oc.Training = parseOperationConfig(training)
+	}
+	if export, ok := m["export"].(map[string]interface{}); ok {
+		oc.Export = parseOperationConfig(export)
+	}
+
+	return oc
+}
+
+func parseOperationConfig(m map[string]interface{}) OperationConfig {
+	oc := OperationConfig{}
+
+	if allowed, ok := m["allowed"].(bool); ok {
+		oc.Allowed = allowed
+	}
+	if requiresCitation, ok := m["requires_citation"].(bool); ok {
+		oc.RequiresCitation = requiresCitation
+	}
+
+	return oc
+}
+
+func parsePIIRuleConfig(m map[string]interface{}) PIIRuleConfig {
+	rc := PIIRuleConfig{
+		Severity: "block", // Default
+	}
+
+	if name, ok := m["name"].(string); ok {
+		rc.Name = name
+	}
+	if ruleType, ok := m["type"].(string); ok {
+		rc.Type = ruleType
+	}
+	if pattern, ok := m["pattern"].(string); ok {
+		rc.Pattern = pattern
+	}
+	if severity, ok := m["severity"].(string); ok {
+		rc.Severity = severity
+	}
+
+	return rc
+}
+
+// LoadDefaults loads a minimal set of default data (for testing or when no YAML exists).
+func (l *Loader) LoadDefaults(ctx context.Context) error {
+	// Create federal policy with KMK
+	federalPolicy, err := l.store.CreatePolicy(ctx, &CreateSourcePolicyRequest{
+		Name: "KMK & Bundesebene",
+	})
+	if err != nil {
+		return fmt.Errorf("failed to create federal policy: %w", err)
+	}
+
+	trustBoost := 0.95
+	legalBasis := "Amtliche Werke (§5 UrhG)"
+	citation := "Quelle: KMK, {title}, {date}"
+
+	_, err = l.store.CreateSource(ctx, &CreateAllowedSourceRequest{
+		PolicyID:         federalPolicy.ID,
+		Domain:           "kmk.org",
+		Name:             "Kultusministerkonferenz",
+		License:          LicenseParagraph5,
+		LegalBasis:       &legalBasis,
+		CitationTemplate: &citation,
+		TrustBoost:       &trustBoost,
+	})
+	if err != nil {
+		return fmt.Errorf("failed to create KMK source: %w", err)
+	}
+
+	// Create default PII rules
+	defaultRules := DefaultPIIRules()
+	for _, rule := range defaultRules {
+		_, err := l.store.CreatePIIRule(ctx, &CreatePIIRuleRequest{
+			Name:     rule.Name,
+			RuleType: PIIRuleType(rule.Type),
+			Pattern:  rule.Pattern,
+			Severity: PIISeverity(rule.Severity),
+		})
+		if err != nil {
+			return fmt.Errorf("failed to create PII rule %s: %w", rule.Name, err)
+		}
+	}
+
+	return nil
+}
+
+// HasData checks if the policy tables already have data.
+func (l *Loader) HasData(ctx context.Context) (bool, error) {
+	policies, _, err := l.store.ListPolicies(ctx, &PolicyListFilter{Limit: 1})
+	if err != nil {
+		return false, err
+	}
+	return len(policies) > 0, nil
+}
+
+// LoadIfEmpty loads data from YAML only if tables are empty.
+func (l *Loader) LoadIfEmpty(ctx context.Context, path string) error {
+	hasData, err := l.HasData(ctx)
+	if err != nil {
+		return err
+	}
+
+	if hasData {
+		return nil // Already has data, skip loading
+	}
+
+	// Check if file exists
+	if _, err := os.Stat(path); os.IsNotExist(err) {
+		// File doesn't exist, load defaults
+		return l.LoadDefaults(ctx)
+	}
+
+	return l.LoadFromFile(ctx, path)
+}
--- a/edu-search-service/internal/policy/models.go
+++ b/edu-search-service/internal/policy/models.go
@@ -0,0 +1,445 @@
+// Package policy provides whitelist-based data source management for the edu-search-service.
+// It implements source policies, operation permissions, PII detection, and audit logging
+// for compliance with German data protection regulations.
+package policy
+
+import (
+	"encoding/json"
+	"time"
+
+	"github.com/google/uuid"
+)
+
+// =============================================================================
+// ENUMS AND CONSTANTS
+// =============================================================================
+
+// Bundesland represents German federal states (2-letter codes).
+type Bundesland string
+
+const (
+	BundeslandBW Bundesland = "BW" // Baden-Wuerttemberg
+	BundeslandBY Bundesland = "BY" // Bayern
+	BundeslandBE Bundesland = "BE" // Berlin
+	BundeslandBB Bundesland = "BB" // Brandenburg
+	BundeslandHB Bundesland = "HB" // Bremen
+	BundeslandHH Bundesland = "HH" // Hamburg
+	BundeslandHE Bundesland = "HE" // Hessen
+	BundeslandMV Bundesland = "MV" // Mecklenburg-Vorpommern
+	BundeslandNI Bundesland = "NI" // Niedersachsen
+	BundeslandNW Bundesland = "NW" // Nordrhein-Westfalen
+	BundeslandRP Bundesland = "RP" // Rheinland-Pfalz
+	BundeslandSL Bundesland = "SL" // Saarland
+	BundeslandSN Bundesland = "SN" // Sachsen
+	BundeslandST Bundesland = "ST" // Sachsen-Anhalt
+	BundeslandSH Bundesland = "SH" // Schleswig-Holstein
+	BundeslandTH Bundesland = "TH" // Thueringen
+)
+
+// ValidBundeslaender contains all valid German federal state codes.
+var ValidBundeslaender = []Bundesland{
+	BundeslandBW, BundeslandBY, BundeslandBE, BundeslandBB,
+	BundeslandHB, BundeslandHH, BundeslandHE, BundeslandMV,
+	BundeslandNI, BundeslandNW, BundeslandRP, BundeslandSL,
+	BundeslandSN, BundeslandST, BundeslandSH, BundeslandTH,
+}
+
+// License represents allowed license types for data sources.
+type License string
+
+const (
+	LicenseDLDEBY20  License = "DL-DE-BY-2.0" // Datenlizenz Deutschland - Namensnennung
+	LicenseCCBY      License = "CC-BY"        // Creative Commons Attribution
+	LicenseCCBYSA    License = "CC-BY-SA"     // Creative Commons Attribution-ShareAlike
+	LicenseCCBYNC    License = "CC-BY-NC"     // Creative Commons Attribution-NonCommercial
+	LicenseCCBYNCSA  License = "CC-BY-NC-SA"  // Creative Commons Attribution-NonCommercial-ShareAlike
+	LicenseCC0       License = "CC0"          // Public Domain
+	LicenseParagraph5 License = "§5 UrhG"     // Amtliche Werke (German Copyright Act)
+	LicenseCustom    License = "Custom"       // Custom license (requires legal basis)
+)
+
+// Operation represents the types of operations that can be performed on data.
+type Operation string
+
+const (
+	OperationLookup   Operation = "lookup"   // Display/Search
+	OperationRAG      Operation = "rag"      // RAG (Retrieval-Augmented Generation)
+	OperationTraining Operation = "training" // Model Training (VERBOTEN by default)
+	OperationExport   Operation = "export"   // Data Export
+)
+
+// ValidOperations contains all valid operation types.
+var ValidOperations = []Operation{
+	OperationLookup,
+	OperationRAG,
+	OperationTraining,
+	OperationExport,
+}
+
+// PIIRuleType represents the type of PII detection rule.
+type PIIRuleType string
+
+const (
+	PIIRuleTypeRegex   PIIRuleType = "regex"   // Regular expression pattern
+	PIIRuleTypeKeyword PIIRuleType = "keyword" // Keyword matching
+)
+
+// PIISeverity represents the severity level of a PII match.
+type PIISeverity string
+
+const (
+	PIISeverityBlock  PIISeverity = "block"  // Block content completely
+	PIISeverityWarn   PIISeverity = "warn"   // Warn but allow
+	PIISeverityRedact PIISeverity = "redact" // Redact matched content
+)
+
+// AuditAction represents the type of action logged in the audit trail.
+type AuditAction string
+
+const (
+	AuditActionCreate     AuditAction = "create"
+	AuditActionUpdate     AuditAction = "update"
+	AuditActionDelete     AuditAction = "delete"
+	AuditActionActivate   AuditAction = "activate"
+	AuditActionDeactivate AuditAction = "deactivate"
+	AuditActionApprove    AuditAction = "approve"
+)
+
+// AuditEntityType represents the type of entity being audited.
+type AuditEntityType string
+
+const (
+	AuditEntitySourcePolicy        AuditEntityType = "source_policy"
+	AuditEntityAllowedSource       AuditEntityType = "allowed_source"
+	AuditEntityOperationPermission AuditEntityType = "operation_permission"
+	AuditEntityPIIRule             AuditEntityType = "pii_rule"
+)
+
+// BlockReason represents the reason why content was blocked.
+type BlockReason string
+
+const (
+	BlockReasonNotWhitelisted    BlockReason = "not_whitelisted"
+	BlockReasonPIIDetected       BlockReason = "pii_detected"
+	BlockReasonTrainingForbidden BlockReason = "training_forbidden"
+	BlockReasonLicenseViolation  BlockReason = "license_violation"
+	BlockReasonManualBlock       BlockReason = "manual_block"
+)
+
+// =============================================================================
+// CORE MODELS
+// =============================================================================
+
+// SourcePolicy represents a versioned policy for data source management.
+// Policies can be scoped to a specific Bundesland or apply federally (bundesland = nil).
+type SourcePolicy struct {
+	ID          uuid.UUID   `json:"id" db:"id"`
+	Version     int         `json:"version" db:"version"`
+	Name        string      `json:"name" db:"name"`
+	Description *string     `json:"description,omitempty" db:"description"`
+	Bundesland  *Bundesland `json:"bundesland,omitempty" db:"bundesland"`
+	IsActive    bool        `json:"is_active" db:"is_active"`
+	CreatedAt   time.Time   `json:"created_at" db:"created_at"`
+	UpdatedAt   time.Time   `json:"updated_at" db:"updated_at"`
+	ApprovedBy  *uuid.UUID  `json:"approved_by,omitempty" db:"approved_by"`
+	ApprovedAt  *time.Time  `json:"approved_at,omitempty" db:"approved_at"`
+
+	// Joined fields (populated by queries)
+	Sources []AllowedSource `json:"sources,omitempty"`
+}
+
+// AllowedSource represents a whitelisted data source with license information.
+type AllowedSource struct {
+	ID               uuid.UUID `json:"id" db:"id"`
+	PolicyID         uuid.UUID `json:"policy_id" db:"policy_id"`
+	Domain           string    `json:"domain" db:"domain"`
+	Name             string    `json:"name" db:"name"`
+	Description      *string   `json:"description,omitempty" db:"description"`
+	License          License   `json:"license" db:"license"`
+	LegalBasis       *string   `json:"legal_basis,omitempty" db:"legal_basis"`
+	CitationTemplate *string   `json:"citation_template,omitempty" db:"citation_template"`
+	TrustBoost       float64   `json:"trust_boost" db:"trust_boost"`
+	IsActive         bool      `json:"is_active" db:"is_active"`
+	CreatedAt        time.Time `json:"created_at" db:"created_at"`
+	UpdatedAt        time.Time `json:"updated_at" db:"updated_at"`
+
+	// Joined fields (populated by queries)
+	Operations []OperationPermission `json:"operations,omitempty"`
+	PolicyName *string               `json:"policy_name,omitempty"`
+}
+
+// OperationPermission represents the permission matrix for a specific source.
+type OperationPermission struct {
+	ID               uuid.UUID `json:"id" db:"id"`
+	SourceID         uuid.UUID `json:"source_id" db:"source_id"`
+	Operation        Operation `json:"operation" db:"operation"`
+	IsAllowed        bool      `json:"is_allowed" db:"is_allowed"`
+	RequiresCitation bool      `json:"requires_citation" db:"requires_citation"`
+	Notes            *string   `json:"notes,omitempty" db:"notes"`
+	CreatedAt        time.Time `json:"created_at" db:"created_at"`
+	UpdatedAt        time.Time `json:"updated_at" db:"updated_at"`
+}
+
+// PIIRule represents a rule for detecting personally identifiable information.
+type PIIRule struct {
+	ID          uuid.UUID   `json:"id" db:"id"`
+	Name        string      `json:"name" db:"name"`
+	Description *string     `json:"description,omitempty" db:"description"`
+	RuleType    PIIRuleType `json:"rule_type" db:"rule_type"`
+	Pattern     string      `json:"pattern" db:"pattern"`
+	Severity    PIISeverity `json:"severity" db:"severity"`
+	IsActive    bool        `json:"is_active" db:"is_active"`
+	CreatedAt   time.Time   `json:"created_at" db:"created_at"`
+	UpdatedAt   time.Time   `json:"updated_at" db:"updated_at"`
+}
+
+// =============================================================================
+// AUDIT AND LOGGING MODELS
+// =============================================================================
+
+// PolicyAuditLog represents an immutable audit log entry for policy changes.
+type PolicyAuditLog struct {
+	ID         uuid.UUID       `json:"id" db:"id"`
+	Action     AuditAction     `json:"action" db:"action"`
+	EntityType AuditEntityType `json:"entity_type" db:"entity_type"`
+	EntityID   *uuid.UUID      `json:"entity_id,omitempty" db:"entity_id"`
+	OldValue   json.RawMessage `json:"old_value,omitempty" db:"old_value"`
+	NewValue   json.RawMessage `json:"new_value,omitempty" db:"new_value"`
+	UserID     *uuid.UUID      `json:"user_id,omitempty" db:"user_id"`
+	UserEmail  *string         `json:"user_email,omitempty" db:"user_email"`
+	IPAddress  *string         `json:"ip_address,omitempty" db:"ip_address"`
+	UserAgent  *string         `json:"user_agent,omitempty" db:"user_agent"`
+	CreatedAt  time.Time       `json:"created_at" db:"created_at"`
+}
+
+// BlockedContentLog represents a log entry for blocked URLs.
+type BlockedContentLog struct {
+	ID            uuid.UUID       `json:"id" db:"id"`
+	URL           string          `json:"url" db:"url"`
+	Domain        string          `json:"domain" db:"domain"`
+	BlockReason   BlockReason     `json:"block_reason" db:"block_reason"`
+	MatchedRuleID *uuid.UUID      `json:"matched_rule_id,omitempty" db:"matched_rule_id"`
+	Details       json.RawMessage `json:"details,omitempty" db:"details"`
+	CreatedAt     time.Time       `json:"created_at" db:"created_at"`
+}
+
+// =============================================================================
+// REQUEST/RESPONSE MODELS
+// =============================================================================
+
+// CreateSourcePolicyRequest represents a request to create a new policy.
+type CreateSourcePolicyRequest struct {
+	Name        string      `json:"name" binding:"required"`
+	Description *string     `json:"description"`
+	Bundesland  *Bundesland `json:"bundesland"`
+}
+
+// UpdateSourcePolicyRequest represents a request to update a policy.
+type UpdateSourcePolicyRequest struct {
+	Name        *string     `json:"name"`
+	Description *string     `json:"description"`
+	Bundesland  *Bundesland `json:"bundesland"`
+	IsActive    *bool       `json:"is_active"`
+}
+
+// CreateAllowedSourceRequest represents a request to create a new allowed source.
+type CreateAllowedSourceRequest struct {
+	PolicyID         uuid.UUID `json:"policy_id" binding:"required"`
+	Domain           string    `json:"domain" binding:"required"`
+	Name             string    `json:"name" binding:"required"`
+	Description      *string   `json:"description"`
+	License          License   `json:"license" binding:"required"`
+	LegalBasis       *string   `json:"legal_basis"`
+	CitationTemplate *string   `json:"citation_template"`
+	TrustBoost       *float64  `json:"trust_boost"`
+}
+
+// UpdateAllowedSourceRequest represents a request to update an allowed source.
+type UpdateAllowedSourceRequest struct {
+	Domain           *string  `json:"domain"`
+	Name             *string  `json:"name"`
+	Description      *string  `json:"description"`
+	License          *License `json:"license"`
+	LegalBasis       *string  `json:"legal_basis"`
+	CitationTemplate *string  `json:"citation_template"`
+	TrustBoost       *float64 `json:"trust_boost"`
+	IsActive         *bool    `json:"is_active"`
+}
+
+// UpdateOperationPermissionRequest represents a request to update operation permissions.
+type UpdateOperationPermissionRequest struct {
+	IsAllowed        *bool   `json:"is_allowed"`
+	RequiresCitation *bool   `json:"requires_citation"`
+	Notes            *string `json:"notes"`
+}
+
+// CreatePIIRuleRequest represents a request to create a new PII rule.
+type CreatePIIRuleRequest struct {
+	Name        string      `json:"name" binding:"required"`
+	Description *string     `json:"description"`
+	RuleType    PIIRuleType `json:"rule_type" binding:"required"`
+	Pattern     string      `json:"pattern" binding:"required"`
+	Severity    PIISeverity `json:"severity"`
+}
+
+// UpdatePIIRuleRequest represents a request to update a PII rule.
+type UpdatePIIRuleRequest struct {
+	Name        *string      `json:"name"`
+	Description *string      `json:"description"`
+	RuleType    *PIIRuleType `json:"rule_type"`
+	Pattern     *string      `json:"pattern"`
+	Severity    *PIISeverity `json:"severity"`
+	IsActive    *bool        `json:"is_active"`
+}
+
+// CheckComplianceRequest represents a request to check URL compliance.
+type CheckComplianceRequest struct {
+	URL        string     `json:"url" binding:"required"`
+	Operation  Operation  `json:"operation" binding:"required"`
+	Bundesland *Bundesland `json:"bundesland"`
+}
+
+// CheckComplianceResponse represents the compliance check result.
+type CheckComplianceResponse struct {
+	IsAllowed        bool     `json:"is_allowed"`
+	Source           *AllowedSource `json:"source,omitempty"`
+	BlockReason      *BlockReason   `json:"block_reason,omitempty"`
+	RequiresCitation bool     `json:"requires_citation"`
+	CitationTemplate *string  `json:"citation_template,omitempty"`
+	License          *License `json:"license,omitempty"`
+}
+
+// PIITestRequest represents a request to test PII detection.
+type PIITestRequest struct {
+	Text string `json:"text" binding:"required"`
+}
+
+// PIIMatch represents a single PII match in text.
+type PIIMatch struct {
+	RuleID     uuid.UUID   `json:"rule_id"`
+	RuleName   string      `json:"rule_name"`
+	RuleType   PIIRuleType `json:"rule_type"`
+	Severity   PIISeverity `json:"severity"`
+	Match      string      `json:"match"`
+	StartIndex int         `json:"start_index"`
+	EndIndex   int         `json:"end_index"`
+}
+
+// PIITestResponse represents the result of PII detection test.
+type PIITestResponse struct {
+	HasPII      bool       `json:"has_pii"`
+	Matches     []PIIMatch `json:"matches"`
+	BlockLevel  PIISeverity `json:"block_level"`
+	ShouldBlock bool       `json:"should_block"`
+}
+
+// =============================================================================
+// LIST/FILTER MODELS
+// =============================================================================
+
+// PolicyListFilter represents filters for listing policies.
+type PolicyListFilter struct {
+	Bundesland *Bundesland `form:"bundesland"`
+	IsActive   *bool       `form:"is_active"`
+	Limit      int         `form:"limit"`
+	Offset     int         `form:"offset"`
+}
+
+// SourceListFilter represents filters for listing sources.
+type SourceListFilter struct {
+	PolicyID   *uuid.UUID `form:"policy_id"`
+	Domain     *string    `form:"domain"`
+	License    *License   `form:"license"`
+	IsActive   *bool      `form:"is_active"`
+	Limit      int        `form:"limit"`
+	Offset     int        `form:"offset"`
+}
+
+// AuditLogFilter represents filters for querying audit logs.
+type AuditLogFilter struct {
+	EntityType *AuditEntityType `form:"entity_type"`
+	EntityID   *uuid.UUID       `form:"entity_id"`
+	Action     *AuditAction     `form:"action"`
+	UserEmail  *string          `form:"user_email"`
+	FromDate   *time.Time       `form:"from"`
+	ToDate     *time.Time       `form:"to"`
+	Limit      int              `form:"limit"`
+	Offset     int              `form:"offset"`
+}
+
+// BlockedContentFilter represents filters for querying blocked content logs.
+type BlockedContentFilter struct {
+	Domain      *string      `form:"domain"`
+	BlockReason *BlockReason `form:"block_reason"`
+	FromDate    *time.Time   `form:"from"`
+	ToDate      *time.Time   `form:"to"`
+	Limit       int          `form:"limit"`
+	Offset      int          `form:"offset"`
+}
+
+// =============================================================================
+// STATISTICS MODELS
+// =============================================================================
+
+// PolicyStats represents aggregated statistics for the policy system.
+type PolicyStats struct {
+	ActivePolicies    int            `json:"active_policies"`
+	TotalSources      int            `json:"total_sources"`
+	ActiveSources     int            `json:"active_sources"`
+	BlockedToday      int            `json:"blocked_today"`
+	BlockedTotal      int            `json:"blocked_total"`
+	PIIRulesActive    int            `json:"pii_rules_active"`
+	SourcesByLicense  map[string]int `json:"sources_by_license"`
+	BlocksByReason    map[string]int `json:"blocks_by_reason"`
+	ComplianceScore   float64        `json:"compliance_score"`
+}
+
+// =============================================================================
+// YAML CONFIGURATION MODELS
+// =============================================================================
+
+// BundeslaenderConfig represents the YAML configuration for initial data loading.
+type BundeslaenderConfig struct {
+	Federal           PolicyConfig           `yaml:"federal"`
+	Bundeslaender     map[string]PolicyConfig `yaml:",inline"`
+	DefaultOperations OperationsConfig       `yaml:"default_operations"`
+	PIIRules          []PIIRuleConfig        `yaml:"pii_rules"`
+}
+
+// PolicyConfig represents a policy configuration in YAML.
+type PolicyConfig struct {
+	Name    string         `yaml:"name"`
+	Sources []SourceConfig `yaml:"sources"`
+}
+
+// SourceConfig represents a source configuration in YAML.
+type SourceConfig struct {
+	Domain           string  `yaml:"domain"`
+	Name             string  `yaml:"name"`
+	License          string  `yaml:"license"`
+	LegalBasis       string  `yaml:"legal_basis,omitempty"`
+	CitationTemplate string  `yaml:"citation_template,omitempty"`
+	TrustBoost       float64 `yaml:"trust_boost,omitempty"`
+}
+
+// OperationsConfig represents default operation permissions in YAML.
+type OperationsConfig struct {
+	Lookup   OperationConfig `yaml:"lookup"`
+	RAG      OperationConfig `yaml:"rag"`
+	Training OperationConfig `yaml:"training"`
+	Export   OperationConfig `yaml:"export"`
+}
+
+// OperationConfig represents a single operation permission in YAML.
+type OperationConfig struct {
+	Allowed          bool `yaml:"allowed"`
+	RequiresCitation bool `yaml:"requires_citation"`
+}
+
+// PIIRuleConfig represents a PII rule configuration in YAML.
+type PIIRuleConfig struct {
+	Name     string `yaml:"name"`
+	Type     string `yaml:"type"`
+	Pattern  string `yaml:"pattern"`
+	Severity string `yaml:"severity"`
+}
--- a/edu-search-service/internal/policy/pii_detector.go
+++ b/edu-search-service/internal/policy/pii_detector.go
@@ -0,0 +1,350 @@
+package policy
+
+import (
+	"context"
+	"regexp"
+	"strings"
+	"sync"
+)
+
+// PIIDetector detects personally identifiable information in text.
+type PIIDetector struct {
+	store         *Store
+	compiledRules map[string]*regexp.Regexp
+	rulesMu       sync.RWMutex
+}
+
+// NewPIIDetector creates a new PIIDetector instance.
+func NewPIIDetector(store *Store) *PIIDetector {
+	return &PIIDetector{
+		store:         store,
+		compiledRules: make(map[string]*regexp.Regexp),
+	}
+}
+
+// Detect scans text for PII patterns and returns all matches.
+func (d *PIIDetector) Detect(ctx context.Context, text string) (*PIITestResponse, error) {
+	rules, err := d.store.ListPIIRules(ctx, true)
+	if err != nil {
+		return nil, err
+	}
+
+	response := &PIITestResponse{
+		HasPII:      false,
+		Matches:     []PIIMatch{},
+		ShouldBlock: false,
+	}
+
+	highestSeverity := PIISeverity("")
+
+	for _, rule := range rules {
+		matches := d.findMatches(text, &rule)
+		if len(matches) > 0 {
+			response.HasPII = true
+			response.Matches = append(response.Matches, matches...)
+
+			// Track highest severity
+			if compareSeverity(rule.Severity, highestSeverity) > 0 {
+				highestSeverity = rule.Severity
+			}
+		}
+	}
+
+	response.BlockLevel = highestSeverity
+	response.ShouldBlock = highestSeverity == PIISeverityBlock
+
+	return response, nil
+}
+
+// findMatches finds all matches for a single rule in the text.
+func (d *PIIDetector) findMatches(text string, rule *PIIRule) []PIIMatch {
+	var matches []PIIMatch
+
+	switch rule.RuleType {
+	case PIIRuleTypeRegex:
+		matches = d.findRegexMatches(text, rule)
+	case PIIRuleTypeKeyword:
+		matches = d.findKeywordMatches(text, rule)
+	}
+
+	return matches
+}
+
+// findRegexMatches finds all regex pattern matches in text.
+func (d *PIIDetector) findRegexMatches(text string, rule *PIIRule) []PIIMatch {
+	re := d.getCompiledRegex(rule.ID.String(), rule.Pattern)
+	if re == nil {
+		return nil
+	}
+
+	var matches []PIIMatch
+	allMatches := re.FindAllStringIndex(text, -1)
+
+	for _, loc := range allMatches {
+		matches = append(matches, PIIMatch{
+			RuleID:     rule.ID,
+			RuleName:   rule.Name,
+			RuleType:   rule.RuleType,
+			Severity:   rule.Severity,
+			Match:      text[loc[0]:loc[1]],
+			StartIndex: loc[0],
+			EndIndex:   loc[1],
+		})
+	}
+
+	return matches
+}
+
+// findKeywordMatches finds all keyword matches in text (case-insensitive).
+func (d *PIIDetector) findKeywordMatches(text string, rule *PIIRule) []PIIMatch {
+	var matches []PIIMatch
+	lowerText := strings.ToLower(text)
+
+	// Split pattern by commas or pipes for multiple keywords
+	keywords := strings.FieldsFunc(rule.Pattern, func(r rune) bool {
+		return r == ',' || r == '|'
+	})
+
+	for _, keyword := range keywords {
+		keyword = strings.TrimSpace(keyword)
+		if keyword == "" {
+			continue
+		}
+
+		lowerKeyword := strings.ToLower(keyword)
+		startIdx := 0
+
+		for {
+			idx := strings.Index(lowerText[startIdx:], lowerKeyword)
+			if idx == -1 {
+				break
+			}
+
+			actualIdx := startIdx + idx
+			matches = append(matches, PIIMatch{
+				RuleID:     rule.ID,
+				RuleName:   rule.Name,
+				RuleType:   rule.RuleType,
+				Severity:   rule.Severity,
+				Match:      text[actualIdx : actualIdx+len(keyword)],
+				StartIndex: actualIdx,
+				EndIndex:   actualIdx + len(keyword),
+			})
+
+			startIdx = actualIdx + len(keyword)
+		}
+	}
+
+	return matches
+}
+
+// getCompiledRegex returns a compiled regex, caching for performance.
+func (d *PIIDetector) getCompiledRegex(ruleID, pattern string) *regexp.Regexp {
+	d.rulesMu.RLock()
+	re, ok := d.compiledRules[ruleID]
+	d.rulesMu.RUnlock()
+
+	if ok {
+		return re
+	}
+
+	// Compile and cache
+	d.rulesMu.Lock()
+	defer d.rulesMu.Unlock()
+
+	// Double-check after acquiring write lock
+	if re, ok = d.compiledRules[ruleID]; ok {
+		return re
+	}
+
+	compiled, err := regexp.Compile(pattern)
+	if err != nil {
+		// Invalid regex - don't cache
+		return nil
+	}
+
+	d.compiledRules[ruleID] = compiled
+	return compiled
+}
+
+// ClearCache clears the compiled regex cache (call after rule updates).
+func (d *PIIDetector) ClearCache() {
+	d.rulesMu.Lock()
+	defer d.rulesMu.Unlock()
+	d.compiledRules = make(map[string]*regexp.Regexp)
+}
+
+// RefreshRules reloads rules and clears the cache.
+func (d *PIIDetector) RefreshRules() {
+	d.ClearCache()
+}
+
+// compareSeverity compares two severity levels.
+// Returns: 1 if a > b, -1 if a < b, 0 if equal.
+func compareSeverity(a, b PIISeverity) int {
+	severityOrder := map[PIISeverity]int{
+		"":                0,
+		PIISeverityWarn:   1,
+		PIISeverityRedact: 2,
+		PIISeverityBlock:  3,
+	}
+
+	aOrder := severityOrder[a]
+	bOrder := severityOrder[b]
+
+	if aOrder > bOrder {
+		return 1
+	} else if aOrder < bOrder {
+		return -1
+	}
+	return 0
+}
+
+// =============================================================================
+// PREDEFINED PII PATTERNS (German Context)
+// =============================================================================
+
+// DefaultPIIRules returns a set of default PII detection rules for German context.
+func DefaultPIIRules() []PIIRuleConfig {
+	return []PIIRuleConfig{
+		// Email Addresses
+		{
+			Name:     "Email Addresses",
+			Type:     "regex",
+			Pattern:  `[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}`,
+			Severity: "block",
+		},
+		// German Phone Numbers
+		{
+			Name:     "German Phone Numbers",
+			Type:     "regex",
+			Pattern:  `(?:\+49|0)[\s.-]?\d{2,4}[\s.-]?\d{3,}[\s.-]?\d{2,}`,
+			Severity: "block",
+		},
+		// German Mobile Numbers
+		{
+			Name:     "German Mobile Numbers",
+			Type:     "regex",
+			Pattern:  `(?:\+49|0)1[567]\d[\s.-]?\d{3,}[\s.-]?\d{2,}`,
+			Severity: "block",
+		},
+		// IBAN (German)
+		{
+			Name:     "German IBAN",
+			Type:     "regex",
+			Pattern:  `DE\d{2}\s?\d{4}\s?\d{4}\s?\d{4}\s?\d{4}\s?\d{2}`,
+			Severity: "block",
+		},
+		// German Social Security Number (Sozialversicherungsnummer)
+		{
+			Name:     "German Social Security Number",
+			Type:     "regex",
+			Pattern:  `\d{2}[0-3]\d[01]\d{2}[A-Z]\d{3}`,
+			Severity: "block",
+		},
+		// German Tax ID (Steuer-ID)
+		{
+			Name:     "German Tax ID",
+			Type:     "regex",
+			Pattern:  `\d{2}\s?\d{3}\s?\d{3}\s?\d{3}`,
+			Severity: "block",
+		},
+		// Credit Card Numbers (Luhn-compatible patterns)
+		{
+			Name:     "Credit Card Numbers",
+			Type:     "regex",
+			Pattern:  `(?:\d{4}[\s.-]?){3}\d{4}`,
+			Severity: "block",
+		},
+		// German Postal Code + City Pattern (potential address)
+		{
+			Name:     "German Address Pattern",
+			Type:     "regex",
+			Pattern:  `\d{5}\s+[A-ZÄÖÜ][a-zäöüß]+`,
+			Severity: "warn",
+		},
+		// Date of Birth Patterns (DD.MM.YYYY)
+		{
+			Name:     "Date of Birth",
+			Type:     "regex",
+			Pattern:  `(?:geboren|geb\.|Geburtsdatum|DoB)[\s:]*\d{1,2}[\./]\d{1,2}[\./]\d{2,4}`,
+			Severity: "warn",
+		},
+		// Personal Names with Titles
+		{
+			Name:     "Personal Names with Titles",
+			Type:     "regex",
+			Pattern:  `(?:Herr|Frau|Dr\.|Prof\.)\s+[A-ZÄÖÜ][a-zäöüß]+\s+[A-ZÄÖÜ][a-zäöüß]+`,
+			Severity: "warn",
+		},
+		// German Health Insurance Number
+		{
+			Name:     "Health Insurance Number",
+			Type:     "regex",
+			Pattern:  `[A-Z]\d{9}`,
+			Severity: "block",
+		},
+		// Vehicle Registration (German)
+		{
+			Name:     "German Vehicle Registration",
+			Type:     "regex",
+			Pattern:  `[A-ZÄÖÜ]{1,3}[\s-]?[A-Z]{1,2}[\s-]?\d{1,4}[HE]?`,
+			Severity: "warn",
+		},
+	}
+}
+
+// =============================================================================
+// REDACTION
+// =============================================================================
+
+// RedactText redacts PII from text based on the matches.
+func (d *PIIDetector) RedactText(text string, matches []PIIMatch) string {
+	if len(matches) == 0 {
+		return text
+	}
+
+	// Sort matches by start index (descending) to replace from end
+	sortedMatches := make([]PIIMatch, len(matches))
+	copy(sortedMatches, matches)
+
+	// Simple bubble sort for small number of matches
+	for i := 0; i < len(sortedMatches)-1; i++ {
+		for j := 0; j < len(sortedMatches)-i-1; j++ {
+			if sortedMatches[j].StartIndex < sortedMatches[j+1].StartIndex {
+				sortedMatches[j], sortedMatches[j+1] = sortedMatches[j+1], sortedMatches[j]
+			}
+		}
+	}
+
+	result := text
+	for _, match := range sortedMatches {
+		if match.Severity == PIISeverityRedact || match.Severity == PIISeverityBlock {
+			replacement := strings.Repeat("*", match.EndIndex-match.StartIndex)
+			result = result[:match.StartIndex] + replacement + result[match.EndIndex:]
+		}
+	}
+
+	return result
+}
+
+// FilterContent filters content based on PII detection.
+// Returns the filtered content and whether it should be blocked.
+func (d *PIIDetector) FilterContent(ctx context.Context, content string) (string, bool, error) {
+	response, err := d.Detect(ctx, content)
+	if err != nil {
+		return content, false, err
+	}
+
+	if !response.HasPII {
+		return content, false, nil
+	}
+
+	if response.ShouldBlock {
+		return "", true, nil
+	}
+
+	// Redact content for warn/redact severity
+	redacted := d.RedactText(content, response.Matches)
+	return redacted, false, nil
+}
--- a/edu-search-service/internal/policy/policy_test.go
+++ b/edu-search-service/internal/policy/policy_test.go
@@ -0,0 +1,489 @@
+package policy
+
+import (
+	"regexp"
+	"testing"
+)
+
+// =============================================================================
+// MODEL TESTS
+// =============================================================================
+
+func TestBundeslandValidation(t *testing.T) {
+	tests := []struct {
+		name     string
+		bl       Bundesland
+		expected bool
+	}{
+		{"valid NI", BundeslandNI, true},
+		{"valid BY", BundeslandBY, true},
+		{"valid BW", BundeslandBW, true},
+		{"valid NW", BundeslandNW, true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			found := false
+			for _, valid := range ValidBundeslaender {
+				if valid == tt.bl {
+					found = true
+					break
+				}
+			}
+			if found != tt.expected {
+				t.Errorf("Expected %v to be valid=%v, got valid=%v", tt.bl, tt.expected, found)
+			}
+		})
+	}
+}
+
+func TestLicenseValues(t *testing.T) {
+	licenses := []License{
+		LicenseDLDEBY20,
+		LicenseCCBY,
+		LicenseCCBYSA,
+		LicenseCC0,
+		LicenseParagraph5,
+	}
+
+	for _, l := range licenses {
+		if l == "" {
+			t.Errorf("License should not be empty")
+		}
+	}
+}
+
+func TestOperationValues(t *testing.T) {
+	if len(ValidOperations) != 4 {
+		t.Errorf("Expected 4 operations, got %d", len(ValidOperations))
+	}
+
+	expectedOps := []Operation{OperationLookup, OperationRAG, OperationTraining, OperationExport}
+	for _, expected := range expectedOps {
+		found := false
+		for _, op := range ValidOperations {
+			if op == expected {
+				found = true
+				break
+			}
+		}
+		if !found {
+			t.Errorf("Expected operation %s not found in ValidOperations", expected)
+		}
+	}
+}
+
+// =============================================================================
+// PII DETECTOR TESTS
+// =============================================================================
+
+func TestPIIDetector_EmailDetection(t *testing.T) {
+	tests := []struct {
+		name     string
+		text     string
+		hasEmail bool
+	}{
+		{"simple email", "Contact: test@example.com", true},
+		{"email with plus", "Email: user+tag@domain.org", true},
+		{"no email", "This is plain text", false},
+		{"partial email", "user@ is not an email", false},
+		{"multiple emails", "Send to a@b.com and x@y.de", true},
+	}
+
+	// Test using regex pattern directly since we don't have a store
+	emailPattern := `[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}`
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Simple test without database
+			rule := &PIIRule{
+				Name:     "Email",
+				RuleType: PIIRuleTypeRegex,
+				Pattern:  emailPattern,
+				Severity: PIISeverityBlock,
+			}
+
+			detector := &PIIDetector{
+				compiledRules: make(map[string]*regexp.Regexp),
+			}
+
+			matches := detector.findMatches(tt.text, rule)
+			hasMatch := len(matches) > 0
+
+			if hasMatch != tt.hasEmail {
+				t.Errorf("Expected hasEmail=%v, got %v for text: %s", tt.hasEmail, hasMatch, tt.text)
+			}
+		})
+	}
+}
+
+func TestPIIDetector_PhoneDetection(t *testing.T) {
+	tests := []struct {
+		name     string
+		text     string
+		hasPhone bool
+	}{
+		{"german mobile", "Call +49 170 1234567", true},
+		{"german landline", "Tel: 030-12345678", true},
+		{"with spaces", "Phone: 0170 123 4567", true},
+		{"no phone", "This is just text", false},
+		{"US format", "Call 555-123-4567", false}, // Should not match German pattern
+	}
+
+	phonePattern := `(?:\+49|0)[\s.-]?\d{2,4}[\s.-]?\d{3,}[\s.-]?\d{2,}`
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rule := &PIIRule{
+				Name:     "Phone",
+				RuleType: PIIRuleTypeRegex,
+				Pattern:  phonePattern,
+				Severity: PIISeverityBlock,
+			}
+
+			detector := &PIIDetector{
+				compiledRules: make(map[string]*regexp.Regexp),
+			}
+
+			matches := detector.findMatches(tt.text, rule)
+			hasMatch := len(matches) > 0
+
+			if hasMatch != tt.hasPhone {
+				t.Errorf("Expected hasPhone=%v, got %v for text: %s", tt.hasPhone, hasMatch, tt.text)
+			}
+		})
+	}
+}
+
+func TestPIIDetector_IBANDetection(t *testing.T) {
+	tests := []struct {
+		name    string
+		text    string
+		hasIBAN bool
+	}{
+		{"valid IBAN", "IBAN: DE89 3704 0044 0532 0130 00", true},
+		{"compact IBAN", "DE89370400440532013000", true},
+		{"no IBAN", "Just a number: 12345678", false},
+		{"partial", "DE12 is not complete", false},
+	}
+
+	ibanPattern := `DE\d{2}\s?\d{4}\s?\d{4}\s?\d{4}\s?\d{4}\s?\d{2}`
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rule := &PIIRule{
+				Name:     "IBAN",
+				RuleType: PIIRuleTypeRegex,
+				Pattern:  ibanPattern,
+				Severity: PIISeverityBlock,
+			}
+
+			detector := &PIIDetector{
+				compiledRules: make(map[string]*regexp.Regexp),
+			}
+
+			matches := detector.findMatches(tt.text, rule)
+			hasMatch := len(matches) > 0
+
+			if hasMatch != tt.hasIBAN {
+				t.Errorf("Expected hasIBAN=%v, got %v for text: %s", tt.hasIBAN, hasMatch, tt.text)
+			}
+		})
+	}
+}
+
+func TestPIIDetector_KeywordMatching(t *testing.T) {
+	tests := []struct {
+		name     string
+		text     string
+		keywords string
+		expected int
+	}{
+		{"single keyword", "The password is secret", "password", 1},
+		{"multiple keywords", "Password and secret", "password,secret", 2},
+		{"case insensitive", "PASSWORD and Secret", "password,secret", 2},
+		{"no match", "This is safe text", "password,secret", 0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rule := &PIIRule{
+				Name:     "Keywords",
+				RuleType: PIIRuleTypeKeyword,
+				Pattern:  tt.keywords,
+				Severity: PIISeverityWarn,
+			}
+
+			detector := &PIIDetector{
+				compiledRules: make(map[string]*regexp.Regexp),
+			}
+
+			matches := detector.findKeywordMatches(tt.text, rule)
+
+			if len(matches) != tt.expected {
+				t.Errorf("Expected %d matches, got %d for text: %s", tt.expected, len(matches), tt.text)
+			}
+		})
+	}
+}
+
+func TestPIIDetector_Redaction(t *testing.T) {
+	detector := &PIIDetector{
+		compiledRules: make(map[string]*regexp.Regexp),
+	}
+
+	tests := []struct {
+		name     string
+		text     string
+		matches  []PIIMatch
+		expected string
+	}{
+		{
+			"single redaction",
+			"Email: test@example.com",
+			[]PIIMatch{{StartIndex: 7, EndIndex: 23, Severity: PIISeverityBlock}},
+			"Email: ****************",
+		},
+		{
+			"no matches",
+			"Plain text",
+			[]PIIMatch{},
+			"Plain text",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := detector.RedactText(tt.text, tt.matches)
+			if result != tt.expected {
+				t.Errorf("Expected '%s', got '%s'", tt.expected, result)
+			}
+		})
+	}
+}
+
+func TestCompareSeverity(t *testing.T) {
+	tests := []struct {
+		a, b     PIISeverity
+		expected int
+	}{
+		{PIISeverityBlock, PIISeverityWarn, 1},
+		{PIISeverityWarn, PIISeverityBlock, -1},
+		{PIISeverityBlock, PIISeverityBlock, 0},
+		{PIISeverityRedact, PIISeverityWarn, 1},
+		{PIISeverityRedact, PIISeverityBlock, -1},
+	}
+
+	for _, tt := range tests {
+		t.Run(string(tt.a)+"_vs_"+string(tt.b), func(t *testing.T) {
+			result := compareSeverity(tt.a, tt.b)
+			if result != tt.expected {
+				t.Errorf("Expected %d, got %d for %s vs %s", tt.expected, result, tt.a, tt.b)
+			}
+		})
+	}
+}
+
+// =============================================================================
+// ENFORCER TESTS
+// =============================================================================
+
+func TestExtractDomain(t *testing.T) {
+	tests := []struct {
+		name     string
+		url      string
+		expected string
+		hasError bool
+	}{
+		{"full URL", "https://www.example.com/path", "example.com", false},
+		{"with port", "http://example.com:8080/path", "example.com", false},
+		{"subdomain", "https://sub.domain.example.com", "sub.domain.example.com", false},
+		{"no scheme", "example.com/path", "example.com", false},
+		{"www prefix", "https://www.test.de", "test.de", false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result, err := extractDomain(tt.url)
+			if tt.hasError && err == nil {
+				t.Error("Expected error, got nil")
+			}
+			if !tt.hasError && err != nil {
+				t.Errorf("Expected no error, got %v", err)
+			}
+			if result != tt.expected {
+				t.Errorf("Expected '%s', got '%s'", tt.expected, result)
+			}
+		})
+	}
+}
+
+// =============================================================================
+// YAML LOADER TESTS
+// =============================================================================
+
+func TestParseYAML(t *testing.T) {
+	yamlData := `
+federal:
+  name: "Test Federal"
+  sources:
+    - domain: "test.gov"
+      name: "Test Source"
+      license: "§5 UrhG"
+      trust_boost: 0.9
+
+NI:
+  name: "Niedersachsen"
+  sources:
+    - domain: "ni.gov"
+      name: "NI Source"
+      license: "DL-DE-BY-2.0"
+
+default_operations:
+  lookup:
+    allowed: true
+    requires_citation: true
+  training:
+    allowed: false
+    requires_citation: false
+
+pii_rules:
+  - name: "Test Rule"
+    type: "regex"
+    pattern: "test.*pattern"
+    severity: "block"
+`
+
+	config, err := ParseYAML([]byte(yamlData))
+	if err != nil {
+		t.Fatalf("Failed to parse YAML: %v", err)
+	}
+
+	// Test federal
+	if config.Federal.Name != "Test Federal" {
+		t.Errorf("Expected federal name 'Test Federal', got '%s'", config.Federal.Name)
+	}
+	if len(config.Federal.Sources) != 1 {
+		t.Errorf("Expected 1 federal source, got %d", len(config.Federal.Sources))
+	}
+	if config.Federal.Sources[0].Domain != "test.gov" {
+		t.Errorf("Expected domain 'test.gov', got '%s'", config.Federal.Sources[0].Domain)
+	}
+	if config.Federal.Sources[0].TrustBoost != 0.9 {
+		t.Errorf("Expected trust_boost 0.9, got %f", config.Federal.Sources[0].TrustBoost)
+	}
+
+	// Test Bundesland
+	if len(config.Bundeslaender) != 1 {
+		t.Errorf("Expected 1 Bundesland, got %d", len(config.Bundeslaender))
+	}
+	ni, ok := config.Bundeslaender["NI"]
+	if !ok {
+		t.Error("Expected NI in Bundeslaender")
+	}
+	if ni.Name != "Niedersachsen" {
+		t.Errorf("Expected name 'Niedersachsen', got '%s'", ni.Name)
+	}
+
+	// Test operations
+	if !config.DefaultOperations.Lookup.Allowed {
+		t.Error("Expected lookup to be allowed")
+	}
+	if config.DefaultOperations.Training.Allowed {
+		t.Error("Expected training to be NOT allowed")
+	}
+
+	// Test PII rules
+	if len(config.PIIRules) != 1 {
+		t.Errorf("Expected 1 PII rule, got %d", len(config.PIIRules))
+	}
+	if config.PIIRules[0].Name != "Test Rule" {
+		t.Errorf("Expected rule name 'Test Rule', got '%s'", config.PIIRules[0].Name)
+	}
+}
+
+// =============================================================================
+// AUDIT TESTS
+// =============================================================================
+
+func TestMaskPII(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{"short", "ab", "****"},
+		{"medium", "test@email.com", "te****om"},
+		{"long", "very-long-email@example.com", "ve****om"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := maskPII(tt.input)
+			if result != tt.expected {
+				t.Errorf("Expected '%s', got '%s'", tt.expected, result)
+			}
+		})
+	}
+}
+
+// =============================================================================
+// DEFAULT PII RULES TEST
+// =============================================================================
+
+func TestDefaultPIIRules(t *testing.T) {
+	rules := DefaultPIIRules()
+
+	if len(rules) == 0 {
+		t.Error("Expected default PII rules, got none")
+	}
+
+	// Check that each rule has required fields
+	for _, rule := range rules {
+		if rule.Name == "" {
+			t.Error("Rule name should not be empty")
+		}
+		if rule.Type == "" {
+			t.Error("Rule type should not be empty")
+		}
+		if rule.Pattern == "" {
+			t.Error("Rule pattern should not be empty")
+		}
+	}
+
+	// Check for email rule
+	hasEmailRule := false
+	for _, rule := range rules {
+		if rule.Name == "Email Addresses" {
+			hasEmailRule = true
+			break
+		}
+	}
+	if !hasEmailRule {
+		t.Error("Expected email addresses rule in defaults")
+	}
+}
+
+// =============================================================================
+// INTEGRATION TEST HELPERS
+// =============================================================================
+
+// TestFilteredURL tests the FilteredURL struct.
+func TestFilteredURL(t *testing.T) {
+	fu := FilteredURL{
+		URL:              "https://example.com",
+		IsAllowed:        true,
+		RequiresCitation: true,
+	}
+
+	if fu.URL != "https://example.com" {
+		t.Error("URL not set correctly")
+	}
+	if !fu.IsAllowed {
+		t.Error("IsAllowed should be true")
+	}
+	if !fu.RequiresCitation {
+		t.Error("RequiresCitation should be true")
+	}
+}
+
--- a/edu-search-service/internal/policy/store.go
+++ b/edu-search-service/internal/policy/store.go
--- a/edu-search-service/internal/publications/crossref_client.go
+++ b/edu-search-service/internal/publications/crossref_client.go
@@ -0,0 +1,369 @@
+package publications
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"strings"
+	"time"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+	"github.com/google/uuid"
+)
+
+// CrossRefClient is a client for the CrossRef API
+type CrossRefClient struct {
+	client    *http.Client
+	baseURL   string
+	userAgent string
+	email     string // For polite pool access
+}
+
+// CrossRefResponse represents the top-level API response
+type CrossRefResponse struct {
+	Status         string         `json:"status"`
+	MessageType    string         `json:"message-type"`
+	MessageVersion string         `json:"message-version"`
+	Message        CrossRefResult `json:"message"`
+}
+
+// CrossRefResult contains the actual results
+type CrossRefResult struct {
+	TotalResults int              `json:"total-results"`
+	Items        []CrossRefWork   `json:"items"`
+	Query        *CrossRefQuery   `json:"query,omitempty"`
+}
+
+// CrossRefQuery contains query info
+type CrossRefQuery struct {
+	StartIndex int `json:"start-index"`
+	SearchTerms string `json:"search-terms"`
+}
+
+// CrossRefWork represents a single work/publication
+type CrossRefWork struct {
+	DOI            string           `json:"DOI"`
+	Title          []string         `json:"title"`
+	ContainerTitle []string         `json:"container-title"`
+	Publisher      string           `json:"publisher"`
+	Type           string           `json:"type"`
+	Author         []CrossRefAuthor `json:"author"`
+	Issued         CrossRefDate     `json:"issued"`
+	PublishedPrint CrossRefDate     `json:"published-print"`
+	Abstract       string           `json:"abstract"`
+	URL            string           `json:"URL"`
+	Link           []CrossRefLink   `json:"link"`
+	Subject        []string         `json:"subject"`
+	ISSN           []string         `json:"ISSN"`
+	ISBN           []string         `json:"ISBN"`
+	IsCitedByCount int              `json:"is-referenced-by-count"`
+}
+
+// CrossRefAuthor represents an author
+type CrossRefAuthor struct {
+	Given       string `json:"given"`
+	Family      string `json:"family"`
+	ORCID       string `json:"ORCID"`
+	Affiliation []struct {
+		Name string `json:"name"`
+	} `json:"affiliation"`
+	Sequence string `json:"sequence"` // "first" or "additional"
+}
+
+// CrossRefDate represents a date
+type CrossRefDate struct {
+	DateParts [][]int `json:"date-parts"`
+}
+
+// CrossRefLink represents a link to the work
+type CrossRefLink struct {
+	URL         string `json:"URL"`
+	ContentType string `json:"content-type"`
+}
+
+// NewCrossRefClient creates a new CrossRef API client
+func NewCrossRefClient(email string) *CrossRefClient {
+	return &CrossRefClient{
+		client: &http.Client{
+			Timeout: 30 * time.Second,
+		},
+		baseURL:   "https://api.crossref.org",
+		userAgent: "BreakPilot-EduBot/1.0 (https://breakpilot.de; mailto:" + email + ")",
+		email:     email,
+	}
+}
+
+// GetWorkByDOI retrieves a work by its DOI
+func (c *CrossRefClient) GetWorkByDOI(ctx context.Context, doi string) (*database.Publication, error) {
+	// Clean DOI
+	doi = strings.TrimSpace(doi)
+	doi = strings.TrimPrefix(doi, "https://doi.org/")
+	doi = strings.TrimPrefix(doi, "http://doi.org/")
+
+	endpoint := fmt.Sprintf("%s/works/%s", c.baseURL, url.PathEscape(doi))
+
+	req, err := http.NewRequestWithContext(ctx, "GET", endpoint, nil)
+	if err != nil {
+		return nil, err
+	}
+
+	req.Header.Set("User-Agent", c.userAgent)
+
+	resp, err := c.client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode == http.StatusNotFound {
+		return nil, fmt.Errorf("DOI not found: %s", doi)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("CrossRef API error: %d", resp.StatusCode)
+	}
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, err
+	}
+
+	var result struct {
+		Status  string       `json:"status"`
+		Message CrossRefWork `json:"message"`
+	}
+
+	if err := json.Unmarshal(body, &result); err != nil {
+		return nil, err
+	}
+
+	return c.convertToPub(&result.Message), nil
+}
+
+// SearchByAuthor searches for publications by author name
+func (c *CrossRefClient) SearchByAuthor(ctx context.Context, authorName string, limit int) ([]*database.Publication, error) {
+	if limit <= 0 {
+		limit = 20
+	}
+
+	endpoint := fmt.Sprintf("%s/works?query.author=%s&rows=%d&sort=published&order=desc",
+		c.baseURL, url.QueryEscape(authorName), limit)
+
+	return c.searchWorks(ctx, endpoint)
+}
+
+// SearchByAffiliation searches for publications by affiliation (university)
+func (c *CrossRefClient) SearchByAffiliation(ctx context.Context, affiliation string, limit int) ([]*database.Publication, error) {
+	if limit <= 0 {
+		limit = 20
+	}
+
+	endpoint := fmt.Sprintf("%s/works?query.affiliation=%s&rows=%d&sort=published&order=desc",
+		c.baseURL, url.QueryEscape(affiliation), limit)
+
+	return c.searchWorks(ctx, endpoint)
+}
+
+// SearchByORCID searches for publications by ORCID
+func (c *CrossRefClient) SearchByORCID(ctx context.Context, orcid string, limit int) ([]*database.Publication, error) {
+	if limit <= 0 {
+		limit = 100
+	}
+
+	// ORCID format: 0000-0000-0000-0000
+	orcid = strings.TrimPrefix(orcid, "https://orcid.org/")
+
+	endpoint := fmt.Sprintf("%s/works?filter=orcid:%s&rows=%d&sort=published&order=desc",
+		c.baseURL, url.QueryEscape(orcid), limit)
+
+	return c.searchWorks(ctx, endpoint)
+}
+
+// SearchByTitle searches for publications by title
+func (c *CrossRefClient) SearchByTitle(ctx context.Context, title string, limit int) ([]*database.Publication, error) {
+	if limit <= 0 {
+		limit = 10
+	}
+
+	endpoint := fmt.Sprintf("%s/works?query.title=%s&rows=%d",
+		c.baseURL, url.QueryEscape(title), limit)
+
+	return c.searchWorks(ctx, endpoint)
+}
+
+// searchWorks performs a generic search
+func (c *CrossRefClient) searchWorks(ctx context.Context, endpoint string) ([]*database.Publication, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", endpoint, nil)
+	if err != nil {
+		return nil, err
+	}
+
+	req.Header.Set("User-Agent", c.userAgent)
+
+	resp, err := c.client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("CrossRef API error: %d", resp.StatusCode)
+	}
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, err
+	}
+
+	var result CrossRefResponse
+	if err := json.Unmarshal(body, &result); err != nil {
+		return nil, err
+	}
+
+	var pubs []*database.Publication
+	for _, work := range result.Message.Items {
+		pubs = append(pubs, c.convertToPub(&work))
+	}
+
+	return pubs, nil
+}
+
+// convertToPub converts a CrossRef work to our Publication model
+func (c *CrossRefClient) convertToPub(work *CrossRefWork) *database.Publication {
+	pub := &database.Publication{
+		ID:            uuid.New(),
+		CitationCount: work.IsCitedByCount,
+		CrawledAt:     time.Now(),
+	}
+
+	// Title
+	if len(work.Title) > 0 {
+		pub.Title = work.Title[0]
+	}
+
+	// DOI
+	if work.DOI != "" {
+		pub.DOI = &work.DOI
+	}
+
+	// URL
+	if work.URL != "" {
+		pub.URL = &work.URL
+	}
+
+	// Abstract (clean HTML)
+	if work.Abstract != "" {
+		abstract := cleanHTML(work.Abstract)
+		pub.Abstract = &abstract
+	}
+
+	// Year
+	if len(work.Issued.DateParts) > 0 && len(work.Issued.DateParts[0]) > 0 {
+		year := work.Issued.DateParts[0][0]
+		pub.Year = &year
+		if len(work.Issued.DateParts[0]) > 1 {
+			month := work.Issued.DateParts[0][1]
+			pub.Month = &month
+		}
+	}
+
+	// Type
+	pubType := mapCrossRefType(work.Type)
+	pub.PubType = &pubType
+
+	// Venue
+	if len(work.ContainerTitle) > 0 {
+		venue := work.ContainerTitle[0]
+		pub.Venue = &venue
+	}
+
+	// Publisher
+	if work.Publisher != "" {
+		pub.Publisher = &work.Publisher
+	}
+
+	// ISBN
+	if len(work.ISBN) > 0 {
+		pub.ISBN = &work.ISBN[0]
+	}
+
+	// ISSN
+	if len(work.ISSN) > 0 {
+		pub.ISSN = &work.ISSN[0]
+	}
+
+	// Keywords/Subjects
+	if len(work.Subject) > 0 {
+		pub.Keywords = work.Subject
+	}
+
+	// PDF URL
+	for _, link := range work.Link {
+		if strings.Contains(link.ContentType, "pdf") {
+			pub.PDFURL = &link.URL
+			break
+		}
+	}
+
+	// Authors
+	var authors []string
+	for _, author := range work.Author {
+		name := strings.TrimSpace(author.Given + " " + author.Family)
+		if name != "" {
+			authors = append(authors, name)
+		}
+	}
+	pub.Authors = authors
+
+	// Source
+	source := "crossref"
+	pub.Source = &source
+
+	// Store raw data
+	rawData, _ := json.Marshal(work)
+	pub.RawData = rawData
+
+	return pub
+}
+
+// mapCrossRefType maps CrossRef types to our types
+func mapCrossRefType(crType string) string {
+	switch crType {
+	case "journal-article":
+		return "journal"
+	case "proceedings-article", "conference-paper":
+		return "conference"
+	case "book":
+		return "book"
+	case "book-chapter":
+		return "book_chapter"
+	case "dissertation":
+		return "thesis"
+	case "posted-content":
+		return "preprint"
+	default:
+		return "other"
+	}
+}
+
+// cleanHTML removes HTML tags from text
+func cleanHTML(html string) string {
+	// Simple HTML tag removal
+	result := html
+	result = strings.ReplaceAll(result, "<jats:p>", "")
+	result = strings.ReplaceAll(result, "</jats:p>", " ")
+	result = strings.ReplaceAll(result, "<jats:italic>", "")
+	result = strings.ReplaceAll(result, "</jats:italic>", "")
+	result = strings.ReplaceAll(result, "<jats:bold>", "")
+	result = strings.ReplaceAll(result, "</jats:bold>", "")
+	result = strings.ReplaceAll(result, "<p>", "")
+	result = strings.ReplaceAll(result, "</p>", " ")
+
+	// Collapse whitespace
+	result = strings.Join(strings.Fields(result), " ")
+
+	return strings.TrimSpace(result)
+}
--- a/edu-search-service/internal/publications/pub_crawler.go
+++ b/edu-search-service/internal/publications/pub_crawler.go
@@ -0,0 +1,268 @@
+package publications
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"sync"
+	"time"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+	"github.com/google/uuid"
+)
+
+// PublicationCrawler crawls publications for university staff
+type PublicationCrawler struct {
+	repo           *database.Repository
+	crossref       *CrossRefClient
+	rateLimit      time.Duration
+	mu             sync.Mutex
+	lastRequest    time.Time
+}
+
+// CrawlResult contains the result of a publication crawl
+type CrawlResult struct {
+	StaffID       uuid.UUID
+	PubsFound     int
+	PubsNew       int
+	PubsUpdated   int
+	Errors        []string
+	Duration      time.Duration
+}
+
+// NewPublicationCrawler creates a new publication crawler
+func NewPublicationCrawler(repo *database.Repository, email string) *PublicationCrawler {
+	return &PublicationCrawler{
+		repo:      repo,
+		crossref:  NewCrossRefClient(email),
+		rateLimit: time.Second, // CrossRef polite pool: 50 req/sec max
+	}
+}
+
+// CrawlForStaff crawls publications for a single staff member
+func (c *PublicationCrawler) CrawlForStaff(ctx context.Context, staff *database.UniversityStaff) (*CrawlResult, error) {
+	start := time.Now()
+	result := &CrawlResult{
+		StaffID: staff.ID,
+	}
+
+	log.Printf("Starting publication crawl for %s", *staff.FullName)
+
+	var pubs []*database.Publication
+
+	// Strategy 1: Search by ORCID (most reliable)
+	if staff.ORCID != nil && *staff.ORCID != "" {
+		c.waitForRateLimit()
+		orcidPubs, err := c.crossref.SearchByORCID(ctx, *staff.ORCID, 100)
+		if err != nil {
+			result.Errors = append(result.Errors, fmt.Sprintf("ORCID search error: %v", err))
+		} else {
+			pubs = append(pubs, orcidPubs...)
+			log.Printf("Found %d publications via ORCID for %s", len(orcidPubs), *staff.FullName)
+		}
+	}
+
+	// Strategy 2: Search by author name
+	if staff.FullName != nil && *staff.FullName != "" {
+		c.waitForRateLimit()
+		namePubs, err := c.crossref.SearchByAuthor(ctx, *staff.FullName, 50)
+		if err != nil {
+			result.Errors = append(result.Errors, fmt.Sprintf("Name search error: %v", err))
+		} else {
+			// Deduplicate
+			for _, pub := range namePubs {
+				if !containsPub(pubs, pub) {
+					pubs = append(pubs, pub)
+				}
+			}
+			log.Printf("Found %d additional publications via name search for %s", len(namePubs), *staff.FullName)
+		}
+	}
+
+	// Save publications and create links
+	for _, pub := range pubs {
+		// Save publication
+		err := c.repo.CreatePublication(ctx, pub)
+		if err != nil {
+			result.Errors = append(result.Errors, fmt.Sprintf("Save error for %s: %v", pub.Title, err))
+			continue
+		}
+
+		result.PubsFound++
+
+		// Link to staff
+		link := &database.StaffPublication{
+			StaffID:       staff.ID,
+			PublicationID: pub.ID,
+		}
+
+		// Determine author position
+		pos := findAuthorPosition(pub, staff)
+		if pos > 0 {
+			link.AuthorPosition = &pos
+		}
+
+		if err := c.repo.LinkStaffPublication(ctx, link); err != nil {
+			result.Errors = append(result.Errors, fmt.Sprintf("Link error: %v", err))
+		}
+	}
+
+	result.Duration = time.Since(start)
+
+	log.Printf("Completed publication crawl for %s: found=%d, duration=%v",
+		*staff.FullName, result.PubsFound, result.Duration)
+
+	return result, nil
+}
+
+// CrawlForUniversity crawls publications for all staff at a university
+func (c *PublicationCrawler) CrawlForUniversity(ctx context.Context, uniID uuid.UUID, limit int) (*database.UniversityCrawlStatus, error) {
+	log.Printf("Starting publication crawl for university %s", uniID)
+
+	// Get staff with ORCID first (more reliable)
+	params := database.StaffSearchParams{
+		UniversityID: &uniID,
+		Limit:        limit,
+	}
+
+	result, err := c.repo.SearchStaff(ctx, params)
+	if err != nil {
+		return nil, err
+	}
+
+	status := &database.UniversityCrawlStatus{
+		UniversityID:     uniID,
+		PubCrawlStatus:   "running",
+	}
+
+	var totalPubs int
+	var errors []string
+
+	for _, staff := range result.Staff {
+		select {
+		case <-ctx.Done():
+			status.PubCrawlStatus = "cancelled"
+			status.PubErrors = append(errors, "Crawl cancelled")
+			return status, ctx.Err()
+		default:
+		}
+
+		crawlResult, err := c.CrawlForStaff(ctx, &staff)
+		if err != nil {
+			errors = append(errors, fmt.Sprintf("%s: %v", staff.LastName, err))
+			continue
+		}
+
+		totalPubs += crawlResult.PubsFound
+		errors = append(errors, crawlResult.Errors...)
+	}
+
+	now := time.Now()
+	status.LastPubCrawl = &now
+	status.PubCrawlStatus = "completed"
+	status.PubCount = totalPubs
+	status.PubErrors = errors
+
+	// Update status in database
+	if err := c.repo.UpdateCrawlStatus(ctx, status); err != nil {
+		log.Printf("Warning: Failed to update crawl status: %v", err)
+	}
+
+	log.Printf("Completed publication crawl for university %s: %d publications found", uniID, totalPubs)
+
+	return status, nil
+}
+
+// ResolveDOI resolves a DOI and saves the publication
+func (c *PublicationCrawler) ResolveDOI(ctx context.Context, doi string) (*database.Publication, error) {
+	c.waitForRateLimit()
+
+	pub, err := c.crossref.GetWorkByDOI(ctx, doi)
+	if err != nil {
+		return nil, err
+	}
+
+	if err := c.repo.CreatePublication(ctx, pub); err != nil {
+		return nil, err
+	}
+
+	return pub, nil
+}
+
+// waitForRateLimit enforces rate limiting
+func (c *PublicationCrawler) waitForRateLimit() {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	elapsed := time.Since(c.lastRequest)
+	if elapsed < c.rateLimit {
+		time.Sleep(c.rateLimit - elapsed)
+	}
+
+	c.lastRequest = time.Now()
+}
+
+// containsPub checks if a publication is already in the list (by DOI or title)
+func containsPub(pubs []*database.Publication, pub *database.Publication) bool {
+	for _, existing := range pubs {
+		// Check DOI
+		if pub.DOI != nil && existing.DOI != nil && *pub.DOI == *existing.DOI {
+			return true
+		}
+		// Check title (rough match)
+		if pub.Title == existing.Title {
+			return true
+		}
+	}
+	return false
+}
+
+// findAuthorPosition finds the position of a staff member in the author list
+func findAuthorPosition(pub *database.Publication, staff *database.UniversityStaff) int {
+	for i, author := range pub.Authors {
+		// Check if author name matches staff
+		if staff.LastName != "" && containsIgnoreCase(author, staff.LastName) {
+			return i + 1
+		}
+	}
+	return 0
+}
+
+// containsIgnoreCase checks if s contains substr (case insensitive)
+func containsIgnoreCase(s, substr string) bool {
+	return len(s) >= len(substr) &&
+		(s == substr ||
+			len(substr) == 0 ||
+			(len(s) > 0 && containsIgnoreCaseHelper(s, substr)))
+}
+
+func containsIgnoreCaseHelper(s, substr string) bool {
+	for i := 0; i <= len(s)-len(substr); i++ {
+		if equalFold(s[i:i+len(substr)], substr) {
+			return true
+		}
+	}
+	return false
+}
+
+func equalFold(s1, s2 string) bool {
+	if len(s1) != len(s2) {
+		return false
+	}
+	for i := 0; i < len(s1); i++ {
+		c1, c2 := s1[i], s2[i]
+		if c1 != c2 {
+			// Simple ASCII case folding
+			if c1 >= 'A' && c1 <= 'Z' {
+				c1 += 'a' - 'A'
+			}
+			if c2 >= 'A' && c2 <= 'Z' {
+				c2 += 'a' - 'A'
+			}
+			if c1 != c2 {
+				return false
+			}
+		}
+	}
+	return true
+}
--- a/edu-search-service/internal/publications/pub_crawler_test.go
+++ b/edu-search-service/internal/publications/pub_crawler_test.go
@@ -0,0 +1,188 @@
+package publications
+
+import (
+	"testing"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+)
+
+func TestContainsPub_ByDOI(t *testing.T) {
+	doi1 := "10.1000/test1"
+	doi2 := "10.1000/test2"
+	doi3 := "10.1000/test3"
+
+	pubs := []*database.Publication{
+		{Title: "Paper 1", DOI: &doi1},
+		{Title: "Paper 2", DOI: &doi2},
+	}
+
+	tests := []struct {
+		name     string
+		pub      *database.Publication
+		expected bool
+	}{
+		{
+			name:     "DOI exists in list",
+			pub:      &database.Publication{Title: "Different Title", DOI: &doi1},
+			expected: true,
+		},
+		{
+			name:     "DOI does not exist",
+			pub:      &database.Publication{Title: "New Paper", DOI: &doi3},
+			expected: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := containsPub(pubs, tt.pub)
+			if result != tt.expected {
+				t.Errorf("Expected %v, got %v", tt.expected, result)
+			}
+		})
+	}
+}
+
+func TestContainsPub_ByTitle(t *testing.T) {
+	pubs := []*database.Publication{
+		{Title: "Machine Learning Applications"},
+		{Title: "Deep Neural Networks"},
+	}
+
+	tests := []struct {
+		name     string
+		pub      *database.Publication
+		expected bool
+	}{
+		{
+			name:     "Title exists in list",
+			pub:      &database.Publication{Title: "Machine Learning Applications"},
+			expected: true,
+		},
+		{
+			name:     "Title does not exist",
+			pub:      &database.Publication{Title: "New Research Paper"},
+			expected: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := containsPub(pubs, tt.pub)
+			if result != tt.expected {
+				t.Errorf("Expected %v, got %v", tt.expected, result)
+			}
+		})
+	}
+}
+
+func TestContainsIgnoreCase(t *testing.T) {
+	tests := []struct {
+		name     string
+		s        string
+		substr   string
+		expected bool
+	}{
+		{"Exact match", "Hello World", "Hello", true},
+		{"Case insensitive", "Hello World", "hello", true},
+		{"Case insensitive uppercase", "HELLO WORLD", "world", true},
+		{"Substring in middle", "The quick brown fox", "brown", true},
+		{"No match", "Hello World", "xyz", false},
+		{"Empty substring", "Hello", "", true},
+		{"Empty string", "", "test", false},
+		{"Both empty", "", "", true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := containsIgnoreCase(tt.s, tt.substr)
+			if result != tt.expected {
+				t.Errorf("containsIgnoreCase(%q, %q) = %v, expected %v",
+					tt.s, tt.substr, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestEqualFold(t *testing.T) {
+	tests := []struct {
+		name     string
+		s1       string
+		s2       string
+		expected bool
+	}{
+		{"Same string", "hello", "hello", true},
+		{"Different case", "Hello", "hello", true},
+		{"All uppercase", "HELLO", "hello", true},
+		{"Mixed case", "HeLLo", "hEllO", true},
+		{"Different strings", "hello", "world", false},
+		{"Different length", "hello", "hi", false},
+		{"Empty strings", "", "", true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := equalFold(tt.s1, tt.s2)
+			if result != tt.expected {
+				t.Errorf("equalFold(%q, %q) = %v, expected %v",
+					tt.s1, tt.s2, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestFindAuthorPosition(t *testing.T) {
+	pub := &database.Publication{
+		Title: "Test Paper",
+		Authors: []string{
+			"John Smith",
+			"Maria Müller",
+			"Hans Weber",
+		},
+	}
+
+	tests := []struct {
+		name     string
+		staff    *database.UniversityStaff
+		expected int
+	}{
+		{
+			name: "First author",
+			staff: &database.UniversityStaff{
+				LastName: "Smith",
+			},
+			expected: 1,
+		},
+		{
+			name: "Second author",
+			staff: &database.UniversityStaff{
+				LastName: "Müller",
+			},
+			expected: 2,
+		},
+		{
+			name: "Third author",
+			staff: &database.UniversityStaff{
+				LastName: "Weber",
+			},
+			expected: 3,
+		},
+		{
+			name: "Author not found",
+			staff: &database.UniversityStaff{
+				LastName: "Unknown",
+			},
+			expected: 0,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := findAuthorPosition(pub, tt.staff)
+			if result != tt.expected {
+				t.Errorf("Expected position %d, got %d for author %s",
+					tt.expected, result, tt.staff.LastName)
+			}
+		})
+	}
+}
--- a/edu-search-service/internal/quality/quality.go
+++ b/edu-search-service/internal/quality/quality.go
@@ -0,0 +1,326 @@
+package quality
+
+import (
+	"regexp"
+	"strings"
+)
+
+// Scorer calculates quality scores for documents
+type Scorer struct {
+	weights Weights
+}
+
+// Weights defines the contribution of each factor to the quality score
+type Weights struct {
+	ContentLength     float64 // 0.20 - longer content often more valuable
+	HeadingStructure  float64 // 0.15 - well-structured documents
+	LinkQuality       float64 // 0.15 - low ad/external link density
+	TextToHTMLRatio   float64 // 0.15 - content-rich pages
+	MetadataPresence  float64 // 0.10 - proper title, description
+	LanguageClarity   float64 // 0.10 - German content, no mixed languages
+	ContentFreshness  float64 // 0.10 - indication of update/recency
+	PDFSpecific       float64 // 0.05 - PDF-specific quality signals
+}
+
+// DefaultWeights returns the default quality score weights
+func DefaultWeights() Weights {
+	return Weights{
+		ContentLength:    0.20,
+		HeadingStructure: 0.15,
+		LinkQuality:      0.15,
+		TextToHTMLRatio:  0.15,
+		MetadataPresence: 0.10,
+		LanguageClarity:  0.10,
+		ContentFreshness: 0.10,
+		PDFSpecific:      0.05,
+	}
+}
+
+// ContentFeatures holds extracted features for quality scoring
+type ContentFeatures struct {
+	ContentLength    int
+	HeadingCount     int
+	HeadingDepth     int      // max heading level depth (h1-h6)
+	LinkDensity      float64
+	AdDensity        float64
+	TextToHTMLRatio  float64
+	HasTitle         bool
+	HasDescription   bool
+	HasCanonical     bool
+	Language         string
+	IsPDF            bool
+	PageCount        int      // for PDFs
+	HasTOC           bool     // table of contents
+	DateIndicators   []string // found date patterns
+}
+
+// Score represents the quality score breakdown
+type Score struct {
+	Total            float64 `json:"total"`
+	ContentLength    float64 `json:"content_length"`
+	HeadingStructure float64 `json:"heading_structure"`
+	LinkQuality      float64 `json:"link_quality"`
+	TextToHTMLRatio  float64 `json:"text_html_ratio"`
+	MetadataPresence float64 `json:"metadata_presence"`
+	LanguageClarity  float64 `json:"language_clarity"`
+	ContentFreshness float64 `json:"content_freshness"`
+	PDFSpecific      float64 `json:"pdf_specific"`
+}
+
+// NewScorer creates a quality scorer with default weights
+func NewScorer() *Scorer {
+	return &Scorer{weights: DefaultWeights()}
+}
+
+// NewScorerWithWeights creates a scorer with custom weights
+func NewScorerWithWeights(w Weights) *Scorer {
+	return &Scorer{weights: w}
+}
+
+// Calculate computes the quality score for given features
+func (s *Scorer) Calculate(features ContentFeatures) Score {
+	score := Score{}
+
+	// 1. Content Length Score (0-1)
+	score.ContentLength = s.calculateContentLengthScore(features.ContentLength)
+
+	// 2. Heading Structure Score (0-1)
+	score.HeadingStructure = s.calculateHeadingScore(features.HeadingCount, features.HeadingDepth, features.HasTOC)
+
+	// 3. Link Quality Score (0-1)
+	score.LinkQuality = s.calculateLinkQualityScore(features.LinkDensity, features.AdDensity)
+
+	// 4. Text to HTML Ratio Score (0-1)
+	score.TextToHTMLRatio = s.calculateTextRatioScore(features.TextToHTMLRatio)
+
+	// 5. Metadata Presence Score (0-1)
+	score.MetadataPresence = s.calculateMetadataScore(features.HasTitle, features.HasDescription, features.HasCanonical)
+
+	// 6. Language Clarity Score (0-1)
+	score.LanguageClarity = s.calculateLanguageScore(features.Language)
+
+	// 7. Content Freshness Score (0-1)
+	score.ContentFreshness = s.calculateFreshnessScore(features.DateIndicators)
+
+	// 8. PDF-Specific Score (0-1)
+	if features.IsPDF {
+		score.PDFSpecific = s.calculatePDFScore(features.PageCount, features.ContentLength)
+	} else {
+		score.PDFSpecific = 1.0 // full score for non-PDFs (no penalty)
+	}
+
+	// Calculate weighted total
+	score.Total = score.ContentLength*s.weights.ContentLength +
+		score.HeadingStructure*s.weights.HeadingStructure +
+		score.LinkQuality*s.weights.LinkQuality +
+		score.TextToHTMLRatio*s.weights.TextToHTMLRatio +
+		score.MetadataPresence*s.weights.MetadataPresence +
+		score.LanguageClarity*s.weights.LanguageClarity +
+		score.ContentFreshness*s.weights.ContentFreshness +
+		score.PDFSpecific*s.weights.PDFSpecific
+
+	// Clamp to 0-1
+	if score.Total > 1.0 {
+		score.Total = 1.0
+	}
+	if score.Total < 0 {
+		score.Total = 0
+	}
+
+	return score
+}
+
+// calculateContentLengthScore scores based on content length
+func (s *Scorer) calculateContentLengthScore(length int) float64 {
+	// Optimal range: 1000-10000 characters
+	// Too short (<500): low quality
+	// Too long (>20000): might be noise/boilerplate
+	switch {
+	case length < 200:
+		return 0.1
+	case length < 500:
+		return 0.3
+	case length < 1000:
+		return 0.6
+	case length < 3000:
+		return 0.8
+	case length < 10000:
+		return 1.0
+	case length < 20000:
+		return 0.9
+	default:
+		return 0.7 // very long documents might have quality issues
+	}
+}
+
+// calculateHeadingScore scores heading structure
+func (s *Scorer) calculateHeadingScore(count, depth int, hasTOC bool) float64 {
+	score := 0.0
+
+	// Headings present
+	if count > 0 {
+		score += 0.4
+	}
+	if count >= 3 {
+		score += 0.2
+	}
+
+	// Depth variety (proper hierarchy)
+	if depth >= 2 {
+		score += 0.2
+	}
+
+	// Table of contents indicates well-structured document
+	if hasTOC {
+		score += 0.2
+	}
+
+	if score > 1.0 {
+		score = 1.0
+	}
+	return score
+}
+
+// calculateLinkQualityScore scores based on link/ad density
+func (s *Scorer) calculateLinkQualityScore(linkDensity, adDensity float64) float64 {
+	score := 1.0
+
+	// High link density is bad
+	if linkDensity > 0.3 {
+		score -= 0.3
+	} else if linkDensity > 0.2 {
+		score -= 0.1
+	}
+
+	// Any ad density is bad
+	if adDensity > 0.1 {
+		score -= 0.4
+	} else if adDensity > 0.05 {
+		score -= 0.2
+	} else if adDensity > 0 {
+		score -= 0.1
+	}
+
+	if score < 0 {
+		score = 0
+	}
+	return score
+}
+
+// calculateTextRatioScore scores text to HTML ratio
+func (s *Scorer) calculateTextRatioScore(ratio float64) float64 {
+	// Good ratio: 0.2-0.6
+	// Too low: too much markup
+	// Too high: might be plain text dump
+	switch {
+	case ratio < 0.1:
+		return 0.3
+	case ratio < 0.2:
+		return 0.6
+	case ratio < 0.6:
+		return 1.0
+	case ratio < 0.8:
+		return 0.8
+	default:
+		return 0.6
+	}
+}
+
+// calculateMetadataScore scores presence of metadata
+func (s *Scorer) calculateMetadataScore(hasTitle, hasDescription, hasCanonical bool) float64 {
+	score := 0.0
+
+	if hasTitle {
+		score += 0.5
+	}
+	if hasDescription {
+		score += 0.3
+	}
+	if hasCanonical {
+		score += 0.2
+	}
+
+	return score
+}
+
+// calculateLanguageScore scores language clarity
+func (s *Scorer) calculateLanguageScore(language string) float64 {
+	switch strings.ToLower(language) {
+	case "de", "german", "deutsch":
+		return 1.0
+	case "en", "english", "englisch":
+		return 0.8 // English is acceptable
+	case "":
+		return 0.5 // unknown
+	default:
+		return 0.3 // other languages
+	}
+}
+
+// calculateFreshnessScore scores content freshness indicators
+func (s *Scorer) calculateFreshnessScore(dateIndicators []string) float64 {
+	if len(dateIndicators) == 0 {
+		return 0.5 // neutral
+	}
+
+	// Check for recent years (2020+)
+	recentYearPattern := regexp.MustCompile(`202[0-5]`)
+	for _, indicator := range dateIndicators {
+		if recentYearPattern.MatchString(indicator) {
+			return 1.0
+		}
+	}
+
+	// Check for 2015-2019
+	modernPattern := regexp.MustCompile(`201[5-9]`)
+	for _, indicator := range dateIndicators {
+		if modernPattern.MatchString(indicator) {
+			return 0.7
+		}
+	}
+
+	// Older content
+	return 0.4
+}
+
+// calculatePDFScore scores PDF-specific quality
+func (s *Scorer) calculatePDFScore(pageCount, contentLength int) float64 {
+	score := 0.5 // base
+
+	// Page count bonus
+	if pageCount > 1 {
+		score += 0.2
+	}
+	if pageCount > 5 {
+		score += 0.1
+	}
+
+	// Text extraction success
+	if contentLength > 100 {
+		score += 0.2
+	}
+
+	if score > 1.0 {
+		score = 1.0
+	}
+	return score
+}
+
+// ExtractDateIndicators finds date patterns in text
+func ExtractDateIndicators(text string) []string {
+	var indicators []string
+
+	// Pattern: DD.MM.YYYY or YYYY-MM-DD
+	datePatterns := []*regexp.Regexp{
+		regexp.MustCompile(`\d{2}\.\d{2}\.\d{4}`),
+		regexp.MustCompile(`\d{4}-\d{2}-\d{2}`),
+		regexp.MustCompile(`\b20[012][0-9]\b`), // years 2000-2029
+	}
+
+	for _, pattern := range datePatterns {
+		matches := pattern.FindAllString(text, 5) // limit matches
+		indicators = append(indicators, matches...)
+	}
+
+	return indicators
+}
--- a/edu-search-service/internal/quality/quality_test.go
+++ b/edu-search-service/internal/quality/quality_test.go
@@ -0,0 +1,333 @@
+package quality
+
+import (
+	"testing"
+)
+
+func TestNewScorer(t *testing.T) {
+	scorer := NewScorer()
+	if scorer == nil {
+		t.Fatal("Expected non-nil scorer")
+	}
+}
+
+func TestNewScorerWithWeights(t *testing.T) {
+	weights := Weights{
+		ContentLength:    0.5,
+		HeadingStructure: 0.5,
+	}
+	scorer := NewScorerWithWeights(weights)
+
+	if scorer.weights.ContentLength != 0.5 {
+		t.Errorf("Expected weight 0.5, got %f", scorer.weights.ContentLength)
+	}
+}
+
+func TestCalculate_HighQualityDocument(t *testing.T) {
+	scorer := NewScorer()
+
+	features := ContentFeatures{
+		ContentLength:   5000,
+		HeadingCount:    5,
+		HeadingDepth:    3,
+		LinkDensity:     0.1,
+		AdDensity:       0,
+		TextToHTMLRatio: 0.4,
+		HasTitle:        true,
+		HasDescription:  true,
+		HasCanonical:    true,
+		Language:        "de",
+		DateIndicators:  []string{"2024-01-15"},
+	}
+
+	score := scorer.Calculate(features)
+
+	if score.Total < 0.8 {
+		t.Errorf("Expected high quality score (>0.8), got %f", score.Total)
+	}
+}
+
+func TestCalculate_LowQualityDocument(t *testing.T) {
+	scorer := NewScorer()
+
+	features := ContentFeatures{
+		ContentLength:   100,
+		HeadingCount:    0,
+		LinkDensity:     0.5,
+		AdDensity:       0.2,
+		TextToHTMLRatio: 0.05,
+		HasTitle:        false,
+		HasDescription:  false,
+		Language:        "",
+	}
+
+	score := scorer.Calculate(features)
+
+	if score.Total > 0.5 {
+		t.Errorf("Expected low quality score (<0.5), got %f", score.Total)
+	}
+}
+
+func TestCalculateContentLengthScore(t *testing.T) {
+	scorer := NewScorer()
+
+	tests := []struct {
+		length   int
+		minScore float64
+		maxScore float64
+	}{
+		{100, 0.0, 0.2},       // very short
+		{500, 0.5, 0.7},       // short-medium
+		{2000, 0.7, 0.9},      // good
+		{5000, 0.9, 1.0},      // optimal
+		{30000, 0.6, 0.8},     // very long
+	}
+
+	for _, tt := range tests {
+		t.Run("", func(t *testing.T) {
+			score := scorer.calculateContentLengthScore(tt.length)
+			if score < tt.minScore || score > tt.maxScore {
+				t.Errorf("Length %d: expected score in [%f, %f], got %f",
+					tt.length, tt.minScore, tt.maxScore, score)
+			}
+		})
+	}
+}
+
+func TestCalculateHeadingScore(t *testing.T) {
+	scorer := NewScorer()
+
+	// No headings
+	score := scorer.calculateHeadingScore(0, 0, false)
+	if score > 0.1 {
+		t.Errorf("Expected low score for no headings, got %f", score)
+	}
+
+	// Good heading structure
+	score = scorer.calculateHeadingScore(5, 3, true)
+	if score < 0.9 {
+		t.Errorf("Expected high score for good headings, got %f", score)
+	}
+}
+
+func TestCalculateLinkQualityScore(t *testing.T) {
+	scorer := NewScorer()
+
+	// Good: low link and ad density
+	score := scorer.calculateLinkQualityScore(0.1, 0)
+	if score < 0.9 {
+		t.Errorf("Expected high score for good link quality, got %f", score)
+	}
+
+	// Bad: high ad density
+	score = scorer.calculateLinkQualityScore(0.1, 0.2)
+	if score > 0.6 {
+		t.Errorf("Expected low score for high ad density, got %f", score)
+	}
+}
+
+func TestCalculateTextRatioScore(t *testing.T) {
+	scorer := NewScorer()
+
+	tests := []struct {
+		ratio    float64
+		minScore float64
+	}{
+		{0.05, 0.0},  // too low
+		{0.3, 0.9},   // optimal
+		{0.9, 0.5},   // too high (plain text dump)
+	}
+
+	for _, tt := range tests {
+		score := scorer.calculateTextRatioScore(tt.ratio)
+		if score < tt.minScore {
+			t.Errorf("Ratio %f: expected score >= %f, got %f", tt.ratio, tt.minScore, score)
+		}
+	}
+}
+
+func TestCalculateMetadataScore(t *testing.T) {
+	scorer := NewScorer()
+
+	// All metadata present
+	score := scorer.calculateMetadataScore(true, true, true)
+	if score != 1.0 {
+		t.Errorf("Expected 1.0 for all metadata, got %f", score)
+	}
+
+	// No metadata
+	score = scorer.calculateMetadataScore(false, false, false)
+	if score != 0.0 {
+		t.Errorf("Expected 0.0 for no metadata, got %f", score)
+	}
+
+	// Only title
+	score = scorer.calculateMetadataScore(true, false, false)
+	if score != 0.5 {
+		t.Errorf("Expected 0.5 for only title, got %f", score)
+	}
+}
+
+func TestCalculateLanguageScore(t *testing.T) {
+	scorer := NewScorer()
+
+	tests := []struct {
+		language string
+		expected float64
+	}{
+		{"de", 1.0},
+		{"german", 1.0},
+		{"en", 0.8},
+		{"", 0.5},
+		{"fr", 0.3},
+	}
+
+	for _, tt := range tests {
+		score := scorer.calculateLanguageScore(tt.language)
+		if score != tt.expected {
+			t.Errorf("Language '%s': expected %f, got %f", tt.language, tt.expected, score)
+		}
+	}
+}
+
+func TestCalculateFreshnessScore(t *testing.T) {
+	scorer := NewScorer()
+
+	// Recent date
+	score := scorer.calculateFreshnessScore([]string{"2024-06-15"})
+	if score < 0.9 {
+		t.Errorf("Expected high score for recent date, got %f", score)
+	}
+
+	// Older date
+	score = scorer.calculateFreshnessScore([]string{"2016-01-01"})
+	if score > 0.8 {
+		t.Errorf("Expected moderate score for 2016, got %f", score)
+	}
+
+	// No date indicators
+	score = scorer.calculateFreshnessScore(nil)
+	if score != 0.5 {
+		t.Errorf("Expected neutral score for no dates, got %f", score)
+	}
+}
+
+func TestCalculatePDFScore(t *testing.T) {
+	scorer := NewScorer()
+
+	// Multi-page PDF with good content
+	score := scorer.calculatePDFScore(10, 5000)
+	if score < 0.8 {
+		t.Errorf("Expected high score for good PDF, got %f", score)
+	}
+
+	// Single page, little content
+	score = scorer.calculatePDFScore(1, 50)
+	if score > 0.6 {
+		t.Errorf("Expected lower score for poor PDF, got %f", score)
+	}
+}
+
+func TestExtractDateIndicators(t *testing.T) {
+	text := "Lehrplan gültig ab 01.08.2023 - Stand: 2024-01-15. Aktualisiert 2024."
+
+	indicators := ExtractDateIndicators(text)
+
+	if len(indicators) == 0 {
+		t.Error("Expected to find date indicators")
+	}
+
+	// Should find at least the year patterns
+	found2024 := false
+	for _, ind := range indicators {
+		if ind == "2024" || ind == "2023" || ind == "2024-01-15" || ind == "01.08.2023" {
+			found2024 = true
+		}
+	}
+
+	if !found2024 {
+		t.Errorf("Expected to find 2024 or 2023, got: %v", indicators)
+	}
+}
+
+func TestExtractDateIndicators_Empty(t *testing.T) {
+	text := "This text has no dates whatsoever."
+
+	indicators := ExtractDateIndicators(text)
+
+	if len(indicators) != 0 {
+		t.Errorf("Expected no indicators, got: %v", indicators)
+	}
+}
+
+func TestCalculate_PDFDocument(t *testing.T) {
+	scorer := NewScorer()
+
+	features := ContentFeatures{
+		ContentLength:   3000,
+		HeadingCount:    3,
+		HeadingDepth:    2,
+		Language:        "de",
+		IsPDF:           true,
+		PageCount:       8,
+		DateIndicators:  []string{"2023"},
+	}
+
+	score := scorer.Calculate(features)
+
+	// PDF with 8 pages and good content should score well
+	if score.PDFSpecific < 0.8 {
+		t.Errorf("Expected good PDF-specific score, got %f", score.PDFSpecific)
+	}
+
+	if score.Total < 0.5 {
+		t.Errorf("Expected reasonable score for PDF, got %f", score.Total)
+	}
+}
+
+func TestCalculate_ScoreClamping(t *testing.T) {
+	scorer := NewScorer()
+
+	// Even with all perfect scores, total should not exceed 1.0
+	features := ContentFeatures{
+		ContentLength:   5000,
+		HeadingCount:    10,
+		HeadingDepth:    4,
+		HasTOC:          true,
+		LinkDensity:     0,
+		AdDensity:       0,
+		TextToHTMLRatio: 0.4,
+		HasTitle:        true,
+		HasDescription:  true,
+		HasCanonical:    true,
+		Language:        "de",
+		DateIndicators:  []string{"2024"},
+	}
+
+	score := scorer.Calculate(features)
+
+	if score.Total > 1.0 {
+		t.Errorf("Score should be clamped to 1.0, got %f", score.Total)
+	}
+	if score.Total < 0 {
+		t.Errorf("Score should not be negative, got %f", score.Total)
+	}
+}
+
+func TestDefaultWeights(t *testing.T) {
+	weights := DefaultWeights()
+
+	// Sum should be approximately 1.0
+	sum := weights.ContentLength +
+		weights.HeadingStructure +
+		weights.LinkQuality +
+		weights.TextToHTMLRatio +
+		weights.MetadataPresence +
+		weights.LanguageClarity +
+		weights.ContentFreshness +
+		weights.PDFSpecific
+
+	if sum < 0.99 || sum > 1.01 {
+		t.Errorf("Default weights should sum to 1.0, got %f", sum)
+	}
+}
--- a/edu-search-service/internal/robots/robots.go
+++ b/edu-search-service/internal/robots/robots.go
@@ -0,0 +1,282 @@
+package robots
+
+import (
+	"bufio"
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"regexp"
+	"strings"
+	"sync"
+	"time"
+)
+
+// Checker handles robots.txt parsing and checking
+type Checker struct {
+	mu        sync.RWMutex
+	cache     map[string]*RobotsData
+	userAgent string
+	client    *http.Client
+	cacheTTL  time.Duration
+}
+
+// RobotsData holds parsed robots.txt data for a host
+type RobotsData struct {
+	DisallowPatterns []string
+	AllowPatterns    []string
+	CrawlDelay       int // seconds
+	FetchedAt        time.Time
+	Error            error
+}
+
+// NewChecker creates a new robots.txt checker
+func NewChecker(userAgent string) *Checker {
+	return &Checker{
+		cache:     make(map[string]*RobotsData),
+		userAgent: userAgent,
+		client: &http.Client{
+			Timeout: 10 * time.Second,
+		},
+		cacheTTL: 24 * time.Hour, // Cache robots.txt for 24 hours
+	}
+}
+
+// IsAllowed checks if a URL is allowed to be crawled
+func (c *Checker) IsAllowed(ctx context.Context, urlStr string) (bool, error) {
+	u, err := url.Parse(urlStr)
+	if err != nil {
+		return false, fmt.Errorf("invalid URL: %w", err)
+	}
+
+	host := u.Host
+	path := u.Path
+	if path == "" {
+		path = "/"
+	}
+
+	// Get or fetch robots.txt
+	robotsData, err := c.getRobotsData(ctx, u.Scheme, host)
+	if err != nil {
+		// If we can't fetch robots.txt, assume allowed (be lenient)
+		return true, nil
+	}
+
+	// If there was an error fetching robots.txt, allow crawling
+	if robotsData.Error != nil {
+		return true, nil
+	}
+
+	// Check allow rules first (they take precedence)
+	for _, pattern := range robotsData.AllowPatterns {
+		if matchPattern(pattern, path) {
+			return true, nil
+		}
+	}
+
+	// Check disallow rules
+	for _, pattern := range robotsData.DisallowPatterns {
+		if matchPattern(pattern, path) {
+			return false, nil
+		}
+	}
+
+	// If no rules match, allow
+	return true, nil
+}
+
+// GetCrawlDelay returns the crawl delay for a host
+func (c *Checker) GetCrawlDelay(ctx context.Context, urlStr string) (int, error) {
+	u, err := url.Parse(urlStr)
+	if err != nil {
+		return 0, err
+	}
+
+	robotsData, err := c.getRobotsData(ctx, u.Scheme, u.Host)
+	if err != nil || robotsData.Error != nil {
+		return 0, nil
+	}
+
+	return robotsData.CrawlDelay, nil
+}
+
+// getRobotsData fetches and caches robots.txt for a host
+func (c *Checker) getRobotsData(ctx context.Context, scheme, host string) (*RobotsData, error) {
+	c.mu.RLock()
+	data, exists := c.cache[host]
+	c.mu.RUnlock()
+
+	// Return cached data if not expired
+	if exists && time.Since(data.FetchedAt) < c.cacheTTL {
+		return data, nil
+	}
+
+	// Fetch robots.txt
+	robotsURL := fmt.Sprintf("%s://%s/robots.txt", scheme, host)
+	data = c.fetchRobots(ctx, robotsURL)
+
+	// Cache the result
+	c.mu.Lock()
+	c.cache[host] = data
+	c.mu.Unlock()
+
+	return data, nil
+}
+
+// fetchRobots fetches and parses robots.txt
+func (c *Checker) fetchRobots(ctx context.Context, robotsURL string) *RobotsData {
+	data := &RobotsData{
+		FetchedAt: time.Now(),
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "GET", robotsURL, nil)
+	if err != nil {
+		data.Error = err
+		return data
+	}
+
+	req.Header.Set("User-Agent", c.userAgent)
+
+	resp, err := c.client.Do(req)
+	if err != nil {
+		data.Error = err
+		return data
+	}
+	defer resp.Body.Close()
+
+	// If robots.txt doesn't exist, allow everything
+	if resp.StatusCode == http.StatusNotFound {
+		return data
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		data.Error = fmt.Errorf("HTTP %d", resp.StatusCode)
+		return data
+	}
+
+	// Parse the robots.txt
+	c.parseRobotsTxt(data, resp.Body)
+
+	return data
+}
+
+// parseRobotsTxt parses robots.txt content
+func (c *Checker) parseRobotsTxt(data *RobotsData, reader io.Reader) {
+	scanner := bufio.NewScanner(reader)
+
+	// Track which user-agent section we're in
+	inRelevantSection := false
+	inWildcardSection := false
+
+	// Normalize our user agent for matching
+	ourAgent := strings.ToLower(c.userAgent)
+
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+
+		// Skip empty lines and comments
+		if line == "" || strings.HasPrefix(line, "#") {
+			continue
+		}
+
+		// Split on first colon
+		parts := strings.SplitN(line, ":", 2)
+		if len(parts) != 2 {
+			continue
+		}
+
+		directive := strings.ToLower(strings.TrimSpace(parts[0]))
+		value := strings.TrimSpace(parts[1])
+
+		// Remove inline comments
+		if idx := strings.Index(value, "#"); idx >= 0 {
+			value = strings.TrimSpace(value[:idx])
+		}
+
+		switch directive {
+		case "user-agent":
+			agent := strings.ToLower(value)
+			if agent == "*" {
+				inWildcardSection = true
+				inRelevantSection = false
+			} else if strings.Contains(ourAgent, agent) || strings.Contains(agent, "breakpilot") || strings.Contains(agent, "edubot") {
+				inRelevantSection = true
+			} else {
+				inRelevantSection = false
+				inWildcardSection = false
+			}
+
+		case "disallow":
+			if value != "" && (inRelevantSection || inWildcardSection) {
+				data.DisallowPatterns = append(data.DisallowPatterns, value)
+			}
+
+		case "allow":
+			if value != "" && (inRelevantSection || inWildcardSection) {
+				data.AllowPatterns = append(data.AllowPatterns, value)
+			}
+
+		case "crawl-delay":
+			if inRelevantSection || inWildcardSection {
+				var delay int
+				fmt.Sscanf(value, "%d", &delay)
+				if delay > 0 {
+					data.CrawlDelay = delay
+				}
+			}
+		}
+	}
+}
+
+// matchPattern matches a URL path against a robots.txt pattern
+func matchPattern(pattern, path string) bool {
+	// Empty pattern matches nothing
+	if pattern == "" {
+		return false
+	}
+
+	// Handle wildcards
+	if strings.Contains(pattern, "*") {
+		// Convert to regex
+		regexPattern := regexp.QuoteMeta(pattern)
+		regexPattern = strings.ReplaceAll(regexPattern, `\*`, ".*")
+
+		// Handle $ at end (exact match)
+		if strings.HasSuffix(regexPattern, `\$`) {
+			regexPattern = strings.TrimSuffix(regexPattern, `\$`) + "$"
+		}
+
+		re, err := regexp.Compile("^" + regexPattern)
+		if err != nil {
+			return false
+		}
+		return re.MatchString(path)
+	}
+
+	// Handle $ (exact end match)
+	if strings.HasSuffix(pattern, "$") {
+		return path == strings.TrimSuffix(pattern, "$")
+	}
+
+	// Simple prefix match
+	return strings.HasPrefix(path, pattern)
+}
+
+// ClearCache clears the robots.txt cache
+func (c *Checker) ClearCache() {
+	c.mu.Lock()
+	c.cache = make(map[string]*RobotsData)
+	c.mu.Unlock()
+}
+
+// CacheStats returns cache statistics
+func (c *Checker) CacheStats() (count int, hosts []string) {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+
+	for host := range c.cache {
+		hosts = append(hosts, host)
+	}
+	return len(c.cache), hosts
+}
--- a/edu-search-service/internal/robots/robots_test.go
+++ b/edu-search-service/internal/robots/robots_test.go
@@ -0,0 +1,324 @@
+package robots
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+func TestNewChecker(t *testing.T) {
+	checker := NewChecker("TestBot/1.0")
+	if checker == nil {
+		t.Fatal("Expected non-nil checker")
+	}
+}
+
+func TestIsAllowed_NoRobots(t *testing.T) {
+	// Server that returns 404 for robots.txt
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+	allowed, err := checker.IsAllowed(context.Background(), server.URL+"/some/page")
+
+	if err != nil {
+		t.Errorf("Unexpected error: %v", err)
+	}
+	if !allowed {
+		t.Error("Should be allowed when robots.txt doesn't exist")
+	}
+}
+
+func TestIsAllowed_AllowAll(t *testing.T) {
+	robotsTxt := `User-agent: *
+Allow: /
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/robots.txt" {
+			w.Write([]byte(robotsTxt))
+			return
+		}
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+	allowed, _ := checker.IsAllowed(context.Background(), server.URL+"/any/path")
+
+	if !allowed {
+		t.Error("Should be allowed with Allow: /")
+	}
+}
+
+func TestIsAllowed_DisallowPath(t *testing.T) {
+	robotsTxt := `User-agent: *
+Disallow: /private/
+Disallow: /admin/
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/robots.txt" {
+			w.Write([]byte(robotsTxt))
+			return
+		}
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+
+	// Should be disallowed
+	allowed, _ := checker.IsAllowed(context.Background(), server.URL+"/private/secret")
+	if allowed {
+		t.Error("/private/secret should be disallowed")
+	}
+
+	allowed, _ = checker.IsAllowed(context.Background(), server.URL+"/admin/users")
+	if allowed {
+		t.Error("/admin/users should be disallowed")
+	}
+
+	// Should be allowed
+	allowed, _ = checker.IsAllowed(context.Background(), server.URL+"/public/page")
+	if !allowed {
+		t.Error("/public/page should be allowed")
+	}
+}
+
+func TestIsAllowed_AllowTakesPrecedence(t *testing.T) {
+	robotsTxt := `User-agent: *
+Disallow: /api/
+Allow: /api/public/
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/robots.txt" {
+			w.Write([]byte(robotsTxt))
+			return
+		}
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+
+	// Allow takes precedence
+	allowed, _ := checker.IsAllowed(context.Background(), server.URL+"/api/public/docs")
+	if !allowed {
+		t.Error("/api/public/docs should be allowed (Allow takes precedence)")
+	}
+
+	// Still disallowed
+	allowed, _ = checker.IsAllowed(context.Background(), server.URL+"/api/internal")
+	if allowed {
+		t.Error("/api/internal should be disallowed")
+	}
+}
+
+func TestIsAllowed_SpecificUserAgent(t *testing.T) {
+	robotsTxt := `User-agent: BadBot
+Disallow: /
+
+User-agent: *
+Allow: /
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/robots.txt" {
+			w.Write([]byte(robotsTxt))
+			return
+		}
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	checker := NewChecker("GoodBot/1.0")
+	allowed, _ := checker.IsAllowed(context.Background(), server.URL+"/page")
+
+	if !allowed {
+		t.Error("GoodBot should be allowed")
+	}
+}
+
+func TestGetCrawlDelay(t *testing.T) {
+	robotsTxt := `User-agent: *
+Crawl-delay: 5
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/robots.txt" {
+			w.Write([]byte(robotsTxt))
+			return
+		}
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+	delay, err := checker.GetCrawlDelay(context.Background(), server.URL+"/page")
+
+	if err != nil {
+		t.Errorf("Unexpected error: %v", err)
+	}
+	if delay != 5 {
+		t.Errorf("Expected delay 5, got %d", delay)
+	}
+}
+
+func TestMatchPattern_Simple(t *testing.T) {
+	tests := []struct {
+		pattern string
+		path    string
+		match   bool
+	}{
+		{"/private/", "/private/secret", true},
+		{"/private/", "/public/", false},
+		{"/", "/anything", true},
+		{"", "/anything", false},
+	}
+
+	for _, tt := range tests {
+		result := matchPattern(tt.pattern, tt.path)
+		if result != tt.match {
+			t.Errorf("Pattern '%s' vs Path '%s': expected %v, got %v",
+				tt.pattern, tt.path, tt.match, result)
+		}
+	}
+}
+
+func TestMatchPattern_Wildcard(t *testing.T) {
+	tests := []struct {
+		pattern string
+		path    string
+		match   bool
+	}{
+		{"/*.pdf", "/document.pdf", true},
+		{"/*.pdf", "/folder/doc.pdf", true},
+		{"/*.pdf", "/document.html", false},
+		{"/dir/*/page", "/dir/sub/page", true},
+		{"/dir/*/page", "/dir/other/page", true},
+	}
+
+	for _, tt := range tests {
+		result := matchPattern(tt.pattern, tt.path)
+		if result != tt.match {
+			t.Errorf("Pattern '%s' vs Path '%s': expected %v, got %v",
+				tt.pattern, tt.path, tt.match, result)
+		}
+	}
+}
+
+func TestMatchPattern_EndAnchor(t *testing.T) {
+	tests := []struct {
+		pattern string
+		path    string
+		match   bool
+	}{
+		{"/exact$", "/exact", true},
+		{"/exact$", "/exactmore", false},
+		{"/exact$", "/exact/more", false},
+	}
+
+	for _, tt := range tests {
+		result := matchPattern(tt.pattern, tt.path)
+		if result != tt.match {
+			t.Errorf("Pattern '%s' vs Path '%s': expected %v, got %v",
+				tt.pattern, tt.path, tt.match, result)
+		}
+	}
+}
+
+func TestCacheStats(t *testing.T) {
+	robotsTxt := `User-agent: *
+Allow: /
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Write([]byte(robotsTxt))
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+
+	// Initially empty
+	count, _ := checker.CacheStats()
+	if count != 0 {
+		t.Errorf("Expected 0 cached entries, got %d", count)
+	}
+
+	// Fetch robots.txt
+	checker.IsAllowed(context.Background(), server.URL+"/page")
+
+	// Should have 1 entry
+	count, hosts := checker.CacheStats()
+	if count != 1 {
+		t.Errorf("Expected 1 cached entry, got %d", count)
+	}
+	if len(hosts) != 1 {
+		t.Errorf("Expected 1 host, got %v", hosts)
+	}
+}
+
+func TestClearCache(t *testing.T) {
+	robotsTxt := `User-agent: *
+Allow: /
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Write([]byte(robotsTxt))
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+
+	// Populate cache
+	checker.IsAllowed(context.Background(), server.URL+"/page")
+
+	count, _ := checker.CacheStats()
+	if count != 1 {
+		t.Errorf("Expected 1 cached entry, got %d", count)
+	}
+
+	// Clear cache
+	checker.ClearCache()
+
+	count, _ = checker.CacheStats()
+	if count != 0 {
+		t.Errorf("Expected 0 cached entries after clear, got %d", count)
+	}
+}
+
+func TestParseRobotsTxt_Comments(t *testing.T) {
+	robotsTxt := `# This is a comment
+User-agent: *
+# Another comment
+Disallow: /private/  # inline comment
+Allow: /public/
+`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/robots.txt" {
+			w.Write([]byte(robotsTxt))
+			return
+		}
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	checker := NewChecker("TestBot/1.0")
+
+	allowed, _ := checker.IsAllowed(context.Background(), server.URL+"/public/page")
+	if !allowed {
+		t.Error("/public/page should be allowed")
+	}
+
+	allowed, _ = checker.IsAllowed(context.Background(), server.URL+"/private/page")
+	if allowed {
+		t.Error("/private/page should be disallowed")
+	}
+}
+
+func TestIsAllowed_InvalidURL(t *testing.T) {
+	checker := NewChecker("TestBot/1.0")
+
+	_, err := checker.IsAllowed(context.Background(), "not a valid url ://")
+	if err == nil {
+		t.Error("Expected error for invalid URL")
+	}
+}
--- a/edu-search-service/internal/scheduler/scheduler.go
+++ b/edu-search-service/internal/scheduler/scheduler.go
@@ -0,0 +1,222 @@
+package scheduler
+
+import (
+	"context"
+	"log"
+	"sync"
+	"time"
+)
+
+// CrawlFunc is the function signature for executing a crawl
+type CrawlFunc func(ctx context.Context) error
+
+// Status represents the current scheduler status
+type Status struct {
+	Enabled       bool      `json:"enabled"`
+	Running       bool      `json:"running"`
+	LastRun       time.Time `json:"last_run,omitempty"`
+	LastRunStatus string    `json:"last_run_status,omitempty"`
+	NextRun       time.Time `json:"next_run,omitempty"`
+	Interval      string    `json:"interval"`
+}
+
+// Scheduler handles automatic crawl scheduling
+type Scheduler struct {
+	mu            sync.RWMutex
+	enabled       bool
+	interval      time.Duration
+	crawlFunc     CrawlFunc
+	running       bool
+	lastRun       time.Time
+	lastRunStatus string
+	stopChan      chan struct{}
+	doneChan      chan struct{}
+}
+
+// Config holds scheduler configuration
+type Config struct {
+	Enabled  bool
+	Interval time.Duration
+}
+
+// NewScheduler creates a new crawler scheduler
+func NewScheduler(cfg Config, crawlFunc CrawlFunc) *Scheduler {
+	return &Scheduler{
+		enabled:   cfg.Enabled,
+		interval:  cfg.Interval,
+		crawlFunc: crawlFunc,
+		stopChan:  make(chan struct{}),
+		doneChan:  make(chan struct{}),
+	}
+}
+
+// Start begins the scheduler loop
+func (s *Scheduler) Start() {
+	if !s.enabled {
+		log.Println("Scheduler is disabled")
+		return
+	}
+
+	log.Printf("Scheduler starting with interval: %v", s.interval)
+
+	go s.run()
+}
+
+// Stop gracefully stops the scheduler
+func (s *Scheduler) Stop() {
+	s.mu.Lock()
+	if !s.enabled {
+		s.mu.Unlock()
+		return
+	}
+	s.mu.Unlock()
+
+	close(s.stopChan)
+	<-s.doneChan
+	log.Println("Scheduler stopped")
+}
+
+// run is the main scheduler loop
+func (s *Scheduler) run() {
+	defer close(s.doneChan)
+
+	// Calculate time until first run
+	// Default: run at 2:00 AM to minimize impact
+	now := time.Now()
+	nextRun := s.calculateNextRun(now)
+
+	log.Printf("Scheduler: first crawl scheduled for %v", nextRun)
+
+	timer := time.NewTimer(time.Until(nextRun))
+	defer timer.Stop()
+
+	for {
+		select {
+		case <-s.stopChan:
+			return
+		case <-timer.C:
+			s.executeCrawl()
+			// Schedule next run
+			nextRun = time.Now().Add(s.interval)
+			timer.Reset(s.interval)
+		}
+	}
+}
+
+// calculateNextRun determines when the next crawl should occur
+func (s *Scheduler) calculateNextRun(from time.Time) time.Time {
+	// If interval is 24h or more, schedule for 2:00 AM
+	if s.interval >= 24*time.Hour {
+		next := time.Date(from.Year(), from.Month(), from.Day(), 2, 0, 0, 0, from.Location())
+		if next.Before(from) || next.Equal(from) {
+			next = next.Add(24 * time.Hour)
+		}
+		return next
+	}
+
+	// For shorter intervals, start immediately
+	return from.Add(1 * time.Minute)
+}
+
+// executeCrawl runs the crawl function
+func (s *Scheduler) executeCrawl() {
+	s.mu.Lock()
+	if s.running {
+		s.mu.Unlock()
+		log.Println("Scheduler: crawl already running, skipping")
+		return
+	}
+	s.running = true
+	s.mu.Unlock()
+
+	log.Println("Scheduler: starting scheduled crawl")
+	startTime := time.Now()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 4*time.Hour)
+	defer cancel()
+
+	err := s.crawlFunc(ctx)
+
+	s.mu.Lock()
+	s.running = false
+	s.lastRun = startTime
+	if err != nil {
+		s.lastRunStatus = "failed: " + err.Error()
+		log.Printf("Scheduler: crawl failed after %v: %v", time.Since(startTime), err)
+	} else {
+		s.lastRunStatus = "success"
+		log.Printf("Scheduler: crawl completed successfully in %v", time.Since(startTime))
+	}
+	s.mu.Unlock()
+}
+
+// TriggerCrawl manually triggers a crawl
+func (s *Scheduler) TriggerCrawl() error {
+	s.mu.Lock()
+	if s.running {
+		s.mu.Unlock()
+		return ErrCrawlAlreadyRunning
+	}
+	s.running = true
+	s.mu.Unlock()
+
+	log.Println("Scheduler: manual crawl triggered")
+
+	go func() {
+		startTime := time.Now()
+		ctx, cancel := context.WithTimeout(context.Background(), 4*time.Hour)
+		defer cancel()
+
+		err := s.crawlFunc(ctx)
+
+		s.mu.Lock()
+		s.running = false
+		s.lastRun = startTime
+		if err != nil {
+			s.lastRunStatus = "failed: " + err.Error()
+			log.Printf("Scheduler: manual crawl failed after %v: %v", time.Since(startTime), err)
+		} else {
+			s.lastRunStatus = "success"
+			log.Printf("Scheduler: manual crawl completed successfully in %v", time.Since(startTime))
+		}
+		s.mu.Unlock()
+	}()
+
+	return nil
+}
+
+// Status returns the current scheduler status
+func (s *Scheduler) Status() Status {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	status := Status{
+		Enabled:       s.enabled,
+		Running:       s.running,
+		LastRun:       s.lastRun,
+		LastRunStatus: s.lastRunStatus,
+		Interval:      s.interval.String(),
+	}
+
+	if s.enabled && !s.lastRun.IsZero() {
+		status.NextRun = s.lastRun.Add(s.interval)
+	}
+
+	return status
+}
+
+// IsRunning returns true if a crawl is currently in progress
+func (s *Scheduler) IsRunning() bool {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	return s.running
+}
+
+// Errors
+type SchedulerError string
+
+func (e SchedulerError) Error() string { return string(e) }
+
+const (
+	ErrCrawlAlreadyRunning = SchedulerError("crawl already running")
+)
--- a/edu-search-service/internal/scheduler/scheduler_test.go
+++ b/edu-search-service/internal/scheduler/scheduler_test.go
@@ -0,0 +1,294 @@
+package scheduler
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestNewScheduler(t *testing.T) {
+	callCount := int32(0)
+	crawlFunc := func(ctx context.Context) error {
+		atomic.AddInt32(&callCount, 1)
+		return nil
+	}
+
+	cfg := Config{
+		Enabled:  true,
+		Interval: 1 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	if scheduler == nil {
+		t.Fatal("Expected non-nil scheduler")
+	}
+
+	if !scheduler.enabled {
+		t.Error("Expected scheduler to be enabled")
+	}
+
+	if scheduler.interval != 1*time.Hour {
+		t.Errorf("Expected interval 1h, got %v", scheduler.interval)
+	}
+}
+
+func TestScheduler_Disabled(t *testing.T) {
+	callCount := int32(0)
+	crawlFunc := func(ctx context.Context) error {
+		atomic.AddInt32(&callCount, 1)
+		return nil
+	}
+
+	cfg := Config{
+		Enabled:  false,
+		Interval: 1 * time.Second,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+	scheduler.Start()
+
+	// Wait a bit - crawl should not run
+	time.Sleep(100 * time.Millisecond)
+
+	if atomic.LoadInt32(&callCount) != 0 {
+		t.Error("Crawl should not run when scheduler is disabled")
+	}
+}
+
+func TestScheduler_TriggerCrawl(t *testing.T) {
+	callCount := int32(0)
+	crawlFunc := func(ctx context.Context) error {
+		atomic.AddInt32(&callCount, 1)
+		time.Sleep(50 * time.Millisecond) // Simulate work
+		return nil
+	}
+
+	cfg := Config{
+		Enabled:  false, // Disabled scheduler, but manual trigger should work
+		Interval: 24 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	// Trigger manual crawl
+	err := scheduler.TriggerCrawl()
+	if err != nil {
+		t.Fatalf("TriggerCrawl failed: %v", err)
+	}
+
+	// Wait for crawl to complete
+	time.Sleep(100 * time.Millisecond)
+
+	if atomic.LoadInt32(&callCount) != 1 {
+		t.Errorf("Expected 1 crawl, got %d", atomic.LoadInt32(&callCount))
+	}
+}
+
+func TestScheduler_TriggerCrawl_AlreadyRunning(t *testing.T) {
+	crawlFunc := func(ctx context.Context) error {
+		time.Sleep(200 * time.Millisecond)
+		return nil
+	}
+
+	cfg := Config{
+		Enabled:  false,
+		Interval: 24 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	// First trigger
+	err := scheduler.TriggerCrawl()
+	if err != nil {
+		t.Fatalf("First TriggerCrawl failed: %v", err)
+	}
+
+	// Wait a bit for crawl to start
+	time.Sleep(10 * time.Millisecond)
+
+	// Second trigger should fail
+	err = scheduler.TriggerCrawl()
+	if err != ErrCrawlAlreadyRunning {
+		t.Errorf("Expected ErrCrawlAlreadyRunning, got %v", err)
+	}
+
+	// Wait for crawl to complete
+	time.Sleep(250 * time.Millisecond)
+
+	// Now trigger should work again
+	err = scheduler.TriggerCrawl()
+	if err != nil {
+		t.Errorf("Third TriggerCrawl should succeed: %v", err)
+	}
+}
+
+func TestScheduler_Status(t *testing.T) {
+	crawlFunc := func(ctx context.Context) error {
+		return nil
+	}
+
+	cfg := Config{
+		Enabled:  true,
+		Interval: 24 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	status := scheduler.Status()
+
+	if !status.Enabled {
+		t.Error("Expected enabled=true")
+	}
+
+	if status.Running {
+		t.Error("Expected running=false initially")
+	}
+
+	if status.Interval != "24h0m0s" {
+		t.Errorf("Expected interval '24h0m0s', got '%s'", status.Interval)
+	}
+}
+
+func TestScheduler_Status_AfterCrawl(t *testing.T) {
+	crawlFunc := func(ctx context.Context) error {
+		return nil
+	}
+
+	cfg := Config{
+		Enabled:  false,
+		Interval: 24 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	// Trigger and wait
+	scheduler.TriggerCrawl()
+	time.Sleep(50 * time.Millisecond)
+
+	status := scheduler.Status()
+
+	if status.LastRun.IsZero() {
+		t.Error("Expected LastRun to be set")
+	}
+
+	if status.LastRunStatus != "success" {
+		t.Errorf("Expected status 'success', got '%s'", status.LastRunStatus)
+	}
+}
+
+func TestScheduler_Status_FailedCrawl(t *testing.T) {
+	crawlFunc := func(ctx context.Context) error {
+		return errors.New("connection failed")
+	}
+
+	cfg := Config{
+		Enabled:  false,
+		Interval: 24 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	// Trigger and wait
+	scheduler.TriggerCrawl()
+	time.Sleep(50 * time.Millisecond)
+
+	status := scheduler.Status()
+
+	if status.LastRunStatus != "failed: connection failed" {
+		t.Errorf("Expected failed status, got '%s'", status.LastRunStatus)
+	}
+}
+
+func TestScheduler_IsRunning(t *testing.T) {
+	crawlFunc := func(ctx context.Context) error {
+		time.Sleep(100 * time.Millisecond)
+		return nil
+	}
+
+	cfg := Config{
+		Enabled:  false,
+		Interval: 24 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	if scheduler.IsRunning() {
+		t.Error("Should not be running initially")
+	}
+
+	scheduler.TriggerCrawl()
+	time.Sleep(10 * time.Millisecond)
+
+	if !scheduler.IsRunning() {
+		t.Error("Should be running after trigger")
+	}
+
+	time.Sleep(150 * time.Millisecond)
+
+	if scheduler.IsRunning() {
+		t.Error("Should not be running after completion")
+	}
+}
+
+func TestScheduler_CalculateNextRun_Daily(t *testing.T) {
+	crawlFunc := func(ctx context.Context) error { return nil }
+
+	cfg := Config{
+		Enabled:  true,
+		Interval: 24 * time.Hour,
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	// Test at 1 AM - should schedule for 2 AM same day
+	from := time.Date(2024, 1, 15, 1, 0, 0, 0, time.UTC)
+	next := scheduler.calculateNextRun(from)
+
+	expectedHour := 2
+	if next.Hour() != expectedHour {
+		t.Errorf("Expected hour %d, got %d", expectedHour, next.Hour())
+	}
+
+	if next.Day() != 15 {
+		t.Errorf("Expected day 15, got %d", next.Day())
+	}
+
+	// Test at 3 AM - should schedule for 2 AM next day
+	from = time.Date(2024, 1, 15, 3, 0, 0, 0, time.UTC)
+	next = scheduler.calculateNextRun(from)
+
+	if next.Day() != 16 {
+		t.Errorf("Expected day 16, got %d", next.Day())
+	}
+}
+
+func TestScheduler_CalculateNextRun_Hourly(t *testing.T) {
+	crawlFunc := func(ctx context.Context) error { return nil }
+
+	cfg := Config{
+		Enabled:  true,
+		Interval: 1 * time.Hour, // Less than 24h
+	}
+
+	scheduler := NewScheduler(cfg, crawlFunc)
+
+	from := time.Date(2024, 1, 15, 10, 30, 0, 0, time.UTC)
+	next := scheduler.calculateNextRun(from)
+
+	// Should start in about 1 minute
+	diff := next.Sub(from)
+	if diff < 30*time.Second || diff > 90*time.Second {
+		t.Errorf("Expected ~1 minute delay for short intervals, got %v", diff)
+	}
+}
+
+func TestSchedulerError(t *testing.T) {
+	err := ErrCrawlAlreadyRunning
+
+	if err.Error() != "crawl already running" {
+		t.Errorf("Unexpected error message: %s", err.Error())
+	}
+}
--- a/edu-search-service/internal/search/search.go
+++ b/edu-search-service/internal/search/search.go
@@ -0,0 +1,592 @@
+package search
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	"github.com/opensearch-project/opensearch-go/v2"
+	"github.com/opensearch-project/opensearch-go/v2/opensearchapi"
+)
+
+// SearchRequest represents an API search request
+type SearchRequest struct {
+	Query   string            `json:"q"`
+	Mode    string            `json:"mode"` // keyword, semantic, hybrid
+	Limit   int               `json:"limit"`
+	Offset  int               `json:"offset"`
+	Filters SearchFilters     `json:"filters"`
+	Rerank  bool              `json:"rerank"`
+	Include SearchInclude     `json:"include"`
+}
+
+// SearchFilters for narrowing results
+type SearchFilters struct {
+	Language       []string `json:"language"`
+	CountryHint    []string `json:"country_hint"`
+	SourceCategory []string `json:"source_category"`
+	DocType        []string `json:"doc_type"`
+	SchoolLevel    []string `json:"school_level"`
+	Subjects       []string `json:"subjects"`
+	State          []string `json:"state"`
+	MinTrustScore  float64  `json:"min_trust_score"`
+	DateFrom       string   `json:"date_from"`
+}
+
+// SearchInclude specifies what to include in response
+type SearchInclude struct {
+	Snippets    bool `json:"snippets"`
+	Highlights  bool `json:"highlights"`
+	ContentText bool `json:"content_text"`
+}
+
+// SearchResult represents a single search result
+type SearchResult struct {
+	DocID       string   `json:"doc_id"`
+	Title       string   `json:"title"`
+	URL         string   `json:"url"`
+	Domain      string   `json:"domain"`
+	Language    string   `json:"language"`
+	DocType     string   `json:"doc_type"`
+	SchoolLevel string   `json:"school_level"`
+	Subjects    []string `json:"subjects"`
+	Scores      Scores   `json:"scores"`
+	Snippet     string   `json:"snippet,omitempty"`
+	Highlights  []string `json:"highlights,omitempty"`
+}
+
+// Scores contains all scoring components
+type Scores struct {
+	BM25     float64 `json:"bm25"`
+	Semantic float64 `json:"semantic"`
+	Rerank   float64 `json:"rerank"`
+	Trust    float64 `json:"trust"`
+	Quality  float64 `json:"quality"`
+	Final    float64 `json:"final"`
+}
+
+// SearchResponse is the API response
+type SearchResponse struct {
+	QueryID    string         `json:"query_id"`
+	Results    []SearchResult `json:"results"`
+	Pagination Pagination     `json:"pagination"`
+}
+
+// Pagination info
+type Pagination struct {
+	Limit         int `json:"limit"`
+	Offset        int `json:"offset"`
+	TotalEstimate int `json:"total_estimate"`
+}
+
+// EmbeddingProvider interface for generating embeddings
+type EmbeddingProvider interface {
+	Embed(ctx context.Context, text string) ([]float32, error)
+	IsEnabled() bool
+	Dimension() int
+}
+
+// Service handles search operations
+type Service struct {
+	client            *opensearch.Client
+	indexName         string
+	embeddingProvider EmbeddingProvider
+	semanticEnabled   bool
+}
+
+// NewService creates a new search service
+func NewService(url, username, password, indexName string) (*Service, error) {
+	cfg := opensearch.Config{
+		Addresses: []string{url},
+		Username:  username,
+		Password:  password,
+	}
+
+	client, err := opensearch.NewClient(cfg)
+	if err != nil {
+		return nil, err
+	}
+
+	return &Service{
+		client:          client,
+		indexName:       indexName,
+		semanticEnabled: false,
+	}, nil
+}
+
+// SetEmbeddingProvider configures the embedding provider for semantic search
+func (s *Service) SetEmbeddingProvider(provider EmbeddingProvider) {
+	if provider != nil && provider.IsEnabled() {
+		s.embeddingProvider = provider
+		s.semanticEnabled = true
+	}
+}
+
+// IsSemanticEnabled returns true if semantic search is available
+func (s *Service) IsSemanticEnabled() bool {
+	return s.semanticEnabled && s.embeddingProvider != nil
+}
+
+// Search performs a search query
+func (s *Service) Search(ctx context.Context, req *SearchRequest) (*SearchResponse, error) {
+	// Determine search mode
+	mode := req.Mode
+	if mode == "" {
+		mode = "keyword" // Default to keyword search
+	}
+
+	// For semantic/hybrid modes, generate query embedding
+	var queryEmbedding []float32
+	var embErr error
+	if (mode == "semantic" || mode == "hybrid") && s.IsSemanticEnabled() {
+		queryEmbedding, embErr = s.embeddingProvider.Embed(ctx, req.Query)
+		if embErr != nil {
+			// Fall back to keyword search if embedding fails
+			mode = "keyword"
+		}
+	} else if mode == "semantic" || mode == "hybrid" {
+		// Semantic requested but not enabled, fall back
+		mode = "keyword"
+	}
+
+	// Build OpenSearch query based on mode
+	var query map[string]interface{}
+	switch mode {
+	case "semantic":
+		query = s.buildSemanticQuery(req, queryEmbedding)
+	case "hybrid":
+		query = s.buildHybridQuery(req, queryEmbedding)
+	default:
+		query = s.buildQuery(req)
+	}
+
+	queryJSON, err := json.Marshal(query)
+	if err != nil {
+		return nil, err
+	}
+
+	searchReq := opensearchapi.SearchRequest{
+		Index: []string{s.indexName},
+		Body:  strings.NewReader(string(queryJSON)),
+	}
+
+	res, err := searchReq.Do(ctx, s.client)
+	if err != nil {
+		return nil, err
+	}
+	defer res.Body.Close()
+
+	// Parse response
+	var osResponse struct {
+		Hits struct {
+			Total struct {
+				Value int `json:"value"`
+			} `json:"total"`
+			Hits []struct {
+				ID     string                 `json:"_id"`
+				Score  float64                `json:"_score"`
+				Source map[string]interface{} `json:"_source"`
+				Highlight map[string][]string `json:"highlight,omitempty"`
+			} `json:"hits"`
+		} `json:"hits"`
+	}
+
+	if err := json.NewDecoder(res.Body).Decode(&osResponse); err != nil {
+		return nil, err
+	}
+
+	// Convert to SearchResults
+	results := make([]SearchResult, 0, len(osResponse.Hits.Hits))
+	for _, hit := range osResponse.Hits.Hits {
+		result := s.hitToResult(hit.Source, hit.Score, hit.Highlight, req.Include)
+		results = append(results, result)
+	}
+
+	return &SearchResponse{
+		QueryID: fmt.Sprintf("q-%d", ctx.Value("request_id")),
+		Results: results,
+		Pagination: Pagination{
+			Limit:         req.Limit,
+			Offset:        req.Offset,
+			TotalEstimate: osResponse.Hits.Total.Value,
+		},
+	}, nil
+}
+
+// buildQuery constructs the OpenSearch query
+func (s *Service) buildQuery(req *SearchRequest) map[string]interface{} {
+	// Main query
+	must := []map[string]interface{}{}
+	filter := []map[string]interface{}{}
+
+	// Text search
+	if req.Query != "" {
+		must = append(must, map[string]interface{}{
+			"multi_match": map[string]interface{}{
+				"query":  req.Query,
+				"fields": []string{"title^3", "content_text"},
+				"type":   "best_fields",
+			},
+		})
+	}
+
+	// Filters
+	if len(req.Filters.Language) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"language": req.Filters.Language},
+		})
+	}
+
+	if len(req.Filters.CountryHint) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"country_hint": req.Filters.CountryHint},
+		})
+	}
+
+	if len(req.Filters.SourceCategory) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"source_category": req.Filters.SourceCategory},
+		})
+	}
+
+	if len(req.Filters.DocType) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"doc_type": req.Filters.DocType},
+		})
+	}
+
+	if len(req.Filters.SchoolLevel) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"school_level": req.Filters.SchoolLevel},
+		})
+	}
+
+	if len(req.Filters.Subjects) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"subjects": req.Filters.Subjects},
+		})
+	}
+
+	if len(req.Filters.State) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"state": req.Filters.State},
+		})
+	}
+
+	if req.Filters.MinTrustScore > 0 {
+		filter = append(filter, map[string]interface{}{
+			"range": map[string]interface{}{
+				"trust_score": map[string]interface{}{"gte": req.Filters.MinTrustScore},
+			},
+		})
+	}
+
+	if req.Filters.DateFrom != "" {
+		filter = append(filter, map[string]interface{}{
+			"range": map[string]interface{}{
+				"fetch_time": map[string]interface{}{"gte": req.Filters.DateFrom},
+			},
+		})
+	}
+
+	// Build bool query
+	boolQuery := map[string]interface{}{}
+	if len(must) > 0 {
+		boolQuery["must"] = must
+	}
+	if len(filter) > 0 {
+		boolQuery["filter"] = filter
+	}
+
+	// Construct full query
+	query := map[string]interface{}{
+		"query": map[string]interface{}{
+			"bool": boolQuery,
+		},
+		"from": req.Offset,
+		"size": req.Limit,
+		"_source": []string{
+			"doc_id", "title", "url", "domain", "language",
+			"doc_type", "school_level", "subjects",
+			"trust_score", "quality_score", "snippet_text",
+		},
+	}
+
+	// Add highlighting if requested
+	if req.Include.Highlights {
+		query["highlight"] = map[string]interface{}{
+			"fields": map[string]interface{}{
+				"title":        map[string]interface{}{},
+				"content_text": map[string]interface{}{"fragment_size": 150, "number_of_fragments": 3},
+			},
+		}
+	}
+
+	// Add function score for trust/quality boosting
+	query["query"] = map[string]interface{}{
+		"function_score": map[string]interface{}{
+			"query": query["query"],
+			"functions": []map[string]interface{}{
+				{
+					"field_value_factor": map[string]interface{}{
+						"field":    "trust_score",
+						"factor":   1.5,
+						"modifier": "sqrt",
+						"missing":  0.5,
+					},
+				},
+				{
+					"field_value_factor": map[string]interface{}{
+						"field":    "quality_score",
+						"factor":   1.0,
+						"modifier": "sqrt",
+						"missing":  0.5,
+					},
+				},
+			},
+			"score_mode": "multiply",
+			"boost_mode": "multiply",
+		},
+	}
+
+	return query
+}
+
+// buildSemanticQuery constructs a pure vector search query using k-NN
+func (s *Service) buildSemanticQuery(req *SearchRequest, embedding []float32) map[string]interface{} {
+	filter := s.buildFilters(req)
+
+	// k-NN query for semantic search
+	knnQuery := map[string]interface{}{
+		"content_embedding": map[string]interface{}{
+			"vector": embedding,
+			"k":      req.Limit + req.Offset, // Get enough results for pagination
+		},
+	}
+
+	// Add filter if present
+	if len(filter) > 0 {
+		knnQuery["content_embedding"].(map[string]interface{})["filter"] = map[string]interface{}{
+			"bool": map[string]interface{}{
+				"filter": filter,
+			},
+		}
+	}
+
+	query := map[string]interface{}{
+		"knn":   knnQuery,
+		"from":  req.Offset,
+		"size":  req.Limit,
+		"_source": []string{
+			"doc_id", "title", "url", "domain", "language",
+			"doc_type", "school_level", "subjects",
+			"trust_score", "quality_score", "snippet_text",
+		},
+	}
+
+	// Add highlighting if requested
+	if req.Include.Highlights {
+		query["highlight"] = map[string]interface{}{
+			"fields": map[string]interface{}{
+				"title":        map[string]interface{}{},
+				"content_text": map[string]interface{}{"fragment_size": 150, "number_of_fragments": 3},
+			},
+		}
+	}
+
+	return query
+}
+
+// buildHybridQuery constructs a combined BM25 + vector search query
+func (s *Service) buildHybridQuery(req *SearchRequest, embedding []float32) map[string]interface{} {
+	filter := s.buildFilters(req)
+
+	// Build the bool query for BM25
+	must := []map[string]interface{}{}
+	if req.Query != "" {
+		must = append(must, map[string]interface{}{
+			"multi_match": map[string]interface{}{
+				"query":  req.Query,
+				"fields": []string{"title^3", "content_text"},
+				"type":   "best_fields",
+			},
+		})
+	}
+
+	boolQuery := map[string]interface{}{}
+	if len(must) > 0 {
+		boolQuery["must"] = must
+	}
+	if len(filter) > 0 {
+		boolQuery["filter"] = filter
+	}
+
+	// Convert embedding to []interface{} for JSON
+	embeddingInterface := make([]interface{}, len(embedding))
+	for i, v := range embedding {
+		embeddingInterface[i] = v
+	}
+
+	// Hybrid query using script_score to combine BM25 and cosine similarity
+	// This is a simpler approach than OpenSearch's neural search plugin
+	query := map[string]interface{}{
+		"query": map[string]interface{}{
+			"script_score": map[string]interface{}{
+				"query": map[string]interface{}{
+					"bool": boolQuery,
+				},
+				"script": map[string]interface{}{
+					"source": "cosineSimilarity(params.query_vector, 'content_embedding') + 1.0 + _score * 0.5",
+					"params": map[string]interface{}{
+						"query_vector": embeddingInterface,
+					},
+				},
+			},
+		},
+		"from": req.Offset,
+		"size": req.Limit,
+		"_source": []string{
+			"doc_id", "title", "url", "domain", "language",
+			"doc_type", "school_level", "subjects",
+			"trust_score", "quality_score", "snippet_text",
+		},
+	}
+
+	// Add highlighting if requested
+	if req.Include.Highlights {
+		query["highlight"] = map[string]interface{}{
+			"fields": map[string]interface{}{
+				"title":        map[string]interface{}{},
+				"content_text": map[string]interface{}{"fragment_size": 150, "number_of_fragments": 3},
+			},
+		}
+	}
+
+	return query
+}
+
+// buildFilters constructs the filter array for queries
+func (s *Service) buildFilters(req *SearchRequest) []map[string]interface{} {
+	filter := []map[string]interface{}{}
+
+	if len(req.Filters.Language) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"language": req.Filters.Language},
+		})
+	}
+
+	if len(req.Filters.CountryHint) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"country_hint": req.Filters.CountryHint},
+		})
+	}
+
+	if len(req.Filters.SourceCategory) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"source_category": req.Filters.SourceCategory},
+		})
+	}
+
+	if len(req.Filters.DocType) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"doc_type": req.Filters.DocType},
+		})
+	}
+
+	if len(req.Filters.SchoolLevel) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"school_level": req.Filters.SchoolLevel},
+		})
+	}
+
+	if len(req.Filters.Subjects) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"subjects": req.Filters.Subjects},
+		})
+	}
+
+	if len(req.Filters.State) > 0 {
+		filter = append(filter, map[string]interface{}{
+			"terms": map[string]interface{}{"state": req.Filters.State},
+		})
+	}
+
+	if req.Filters.MinTrustScore > 0 {
+		filter = append(filter, map[string]interface{}{
+			"range": map[string]interface{}{
+				"trust_score": map[string]interface{}{"gte": req.Filters.MinTrustScore},
+			},
+		})
+	}
+
+	if req.Filters.DateFrom != "" {
+		filter = append(filter, map[string]interface{}{
+			"range": map[string]interface{}{
+				"fetch_time": map[string]interface{}{"gte": req.Filters.DateFrom},
+			},
+		})
+	}
+
+	return filter
+}
+
+// hitToResult converts an OpenSearch hit to SearchResult
+func (s *Service) hitToResult(source map[string]interface{}, score float64, highlight map[string][]string, include SearchInclude) SearchResult {
+	result := SearchResult{
+		DocID:       getString(source, "doc_id"),
+		Title:       getString(source, "title"),
+		URL:         getString(source, "url"),
+		Domain:      getString(source, "domain"),
+		Language:    getString(source, "language"),
+		DocType:     getString(source, "doc_type"),
+		SchoolLevel: getString(source, "school_level"),
+		Subjects:    getStringArray(source, "subjects"),
+		Scores: Scores{
+			BM25:    score,
+			Trust:   getFloat(source, "trust_score"),
+			Quality: getFloat(source, "quality_score"),
+			Final:   score, // MVP: final = BM25 * trust * quality (via function_score)
+		},
+	}
+
+	if include.Snippets {
+		result.Snippet = getString(source, "snippet_text")
+	}
+
+	if include.Highlights && highlight != nil {
+		if h, ok := highlight["content_text"]; ok {
+			result.Highlights = h
+		}
+	}
+
+	return result
+}
+
+// Helper functions
+func getString(m map[string]interface{}, key string) string {
+	if v, ok := m[key].(string); ok {
+		return v
+	}
+	return ""
+}
+
+func getFloat(m map[string]interface{}, key string) float64 {
+	if v, ok := m[key].(float64); ok {
+		return v
+	}
+	return 0.0
+}
+
+func getStringArray(m map[string]interface{}, key string) []string {
+	if v, ok := m[key].([]interface{}); ok {
+		result := make([]string, 0, len(v))
+		for _, item := range v {
+			if s, ok := item.(string); ok {
+				result = append(result, s)
+			}
+		}
+		return result
+	}
+	return nil
+}
--- a/edu-search-service/internal/staff/orchestrator_adapter.go
+++ b/edu-search-service/internal/staff/orchestrator_adapter.go
@@ -0,0 +1,217 @@
+// Package staff provides university staff crawling functionality
+package staff
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"time"
+
+	"github.com/google/uuid"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+	"github.com/breakpilot/edu-search-service/internal/orchestrator"
+)
+
+// OrchestratorAdapter adapts the StaffCrawler to the orchestrator.StaffCrawlerInterface
+// This bridges the gap between the generic StaffCrawler and the multi-phase orchestrator
+type OrchestratorAdapter struct {
+	crawler *StaffCrawler
+	repo    *database.Repository
+}
+
+// NewOrchestratorAdapter creates a new adapter that connects StaffCrawler to the orchestrator
+func NewOrchestratorAdapter(crawler *StaffCrawler, repo *database.Repository) *OrchestratorAdapter {
+	return &OrchestratorAdapter{
+		crawler: crawler,
+		repo:    repo,
+	}
+}
+
+// DiscoverSampleProfessor finds at least one professor to validate crawling works for this university
+// This is Phase 1: Quick validation that the university website is crawlable
+func (a *OrchestratorAdapter) DiscoverSampleProfessor(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	start := time.Now()
+	progress := &orchestrator.CrawlProgress{
+		Phase:     orchestrator.PhaseDiscovery,
+		StartedAt: start,
+	}
+
+	log.Printf("[OrchestratorAdapter] Discovery phase for university %s", universityID)
+
+	// Get university from database
+	uni, err := a.repo.GetUniversityByID(ctx, universityID)
+	if err != nil {
+		progress.Errors = append(progress.Errors, fmt.Sprintf("Failed to get university: %v", err))
+		return progress, fmt.Errorf("failed to get university: %w", err)
+	}
+
+	if uni == nil {
+		progress.Errors = append(progress.Errors, "University not found")
+		return progress, fmt.Errorf("university not found: %s", universityID)
+	}
+
+	log.Printf("[OrchestratorAdapter] Discovering staff pages for %s (%s)", uni.Name, uni.URL)
+
+	// Use the crawler to find staff pages (discovery phase)
+	staffPages, err := a.crawler.findStaffPages(ctx, uni)
+	if err != nil {
+		progress.Errors = append(progress.Errors, fmt.Sprintf("Failed to find staff pages: %v", err))
+		return progress, fmt.Errorf("failed to find staff pages: %w", err)
+	}
+
+	log.Printf("[OrchestratorAdapter] Found %d staff pages for %s", len(staffPages), uni.Name)
+
+	// Try to extract at least one professor as validation
+	var sampleFound int
+	for _, pageURL := range staffPages {
+		if sampleFound > 0 {
+			break // We just need to validate one works
+		}
+
+		staffMembers, err := a.crawler.extractStaffFromPage(ctx, pageURL, uni)
+		if err != nil {
+			log.Printf("[OrchestratorAdapter] Error extracting from %s: %v", pageURL, err)
+			continue
+		}
+
+		// Count professors found
+		for _, staff := range staffMembers {
+			if staff.IsProfessor {
+				sampleFound++
+				log.Printf("[OrchestratorAdapter] Found sample professor: %s %s",
+					stringValue(staff.FirstName), staff.LastName)
+				break
+			}
+		}
+
+		// Even non-professors validate the crawler works
+		if sampleFound == 0 && len(staffMembers) > 0 {
+			sampleFound = 1
+			log.Printf("[OrchestratorAdapter] Found sample staff member (not professor): %s %s",
+				stringValue(staffMembers[0].FirstName), staffMembers[0].LastName)
+		}
+	}
+
+	progress.ItemsFound = len(staffPages) // Number of crawlable pages found
+	now := time.Now()
+	progress.CompletedAt = &now
+
+	if sampleFound == 0 && len(staffPages) > 0 {
+		// Pages found but no staff extracted - still consider it successful
+		log.Printf("[OrchestratorAdapter] Discovery completed: %d pages found, extraction may need tuning", len(staffPages))
+	} else if sampleFound == 0 {
+		progress.Errors = append(progress.Errors, "No staff pages found")
+		return progress, fmt.Errorf("no staff pages found for %s", uni.Name)
+	}
+
+	log.Printf("[OrchestratorAdapter] Discovery completed for %s: %d pages found", uni.Name, len(staffPages))
+	return progress, nil
+}
+
+// CrawlProfessors crawls all professors at a university
+// This is Phase 2: Focus on finding professors specifically
+func (a *OrchestratorAdapter) CrawlProfessors(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	start := time.Now()
+	progress := &orchestrator.CrawlProgress{
+		Phase:     orchestrator.PhaseProfessors,
+		StartedAt: start,
+	}
+
+	log.Printf("[OrchestratorAdapter] Professors phase for university %s", universityID)
+
+	// Get university
+	uni, err := a.repo.GetUniversityByID(ctx, universityID)
+	if err != nil || uni == nil {
+		progress.Errors = append(progress.Errors, fmt.Sprintf("Failed to get university: %v", err))
+		return progress, fmt.Errorf("failed to get university: %w", err)
+	}
+
+	// Perform full crawl
+	result, err := a.crawler.CrawlUniversity(ctx, uni)
+	if err != nil {
+		progress.Errors = append(progress.Errors, fmt.Sprintf("Crawl failed: %v", err))
+		return progress, err
+	}
+
+	// Count professors specifically
+	professorCount := 0
+	staffList, err := a.repo.SearchStaff(ctx, database.StaffSearchParams{
+		UniversityID: &universityID,
+		IsProfessor:  boolPtr(true),
+		Limit:        10000,
+	})
+	if err == nil {
+		professorCount = staffList.Total
+	}
+
+	progress.ItemsFound = professorCount
+	progress.ItemsProcessed = result.StaffFound
+	progress.Errors = result.Errors
+	now := time.Now()
+	progress.CompletedAt = &now
+
+	log.Printf("[OrchestratorAdapter] Professors phase completed for %s: %d professors found", uni.Name, professorCount)
+	return progress, nil
+}
+
+// CrawlAllStaff crawls all staff members at a university
+// This is Phase 3: Get all staff (already done in Phase 2, but we verify/extend)
+func (a *OrchestratorAdapter) CrawlAllStaff(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	start := time.Now()
+	progress := &orchestrator.CrawlProgress{
+		Phase:     orchestrator.PhaseAllStaff,
+		StartedAt: start,
+	}
+
+	log.Printf("[OrchestratorAdapter] All Staff phase for university %s", universityID)
+
+	// Get university
+	uni, err := a.repo.GetUniversityByID(ctx, universityID)
+	if err != nil || uni == nil {
+		progress.Errors = append(progress.Errors, fmt.Sprintf("Failed to get university: %v", err))
+		return progress, fmt.Errorf("failed to get university: %w", err)
+	}
+
+	// Run another crawl pass to catch any missed staff
+	result, err := a.crawler.CrawlUniversity(ctx, uni)
+	if err != nil {
+		progress.Errors = result.Errors
+		// Don't fail completely - we may have some staff already
+		log.Printf("[OrchestratorAdapter] All Staff crawl had errors: %v", err)
+	}
+
+	// Get total staff count
+	staffCount := 0
+	staffList, err := a.repo.SearchStaff(ctx, database.StaffSearchParams{
+		UniversityID: &universityID,
+		Limit:        1, // Just need count
+	})
+	if err == nil {
+		staffCount = staffList.Total
+	}
+
+	progress.ItemsFound = staffCount
+	if result != nil {
+		progress.ItemsProcessed = result.StaffFound
+		progress.Errors = result.Errors
+	}
+	now := time.Now()
+	progress.CompletedAt = &now
+
+	log.Printf("[OrchestratorAdapter] All Staff phase completed for %s: %d total staff", uni.Name, staffCount)
+	return progress, nil
+}
+
+// Helper functions
+
+func stringValue(s *string) string {
+	if s == nil {
+		return ""
+	}
+	return *s
+}
+
+func boolPtr(b bool) *bool {
+	return &b
+}
--- a/edu-search-service/internal/staff/patterns.go
+++ b/edu-search-service/internal/staff/patterns.go
@@ -0,0 +1,342 @@
+package staff
+
+import (
+	"regexp"
+	"strings"
+)
+
+// UniversityPatterns contains URL patterns for specific universities
+type UniversityPatterns struct {
+	patterns map[string]UniversityConfig
+}
+
+// UniversityConfig contains crawling configuration for a specific university
+type UniversityConfig struct {
+	StaffListURLs    []string          // URLs to staff listing pages
+	StaffLinkPattern *regexp.Regexp    // Pattern to identify staff profile links
+	NameSelector     string            // CSS selector for person name
+	PositionSelector string            // CSS selector for position
+	EmailSelector    string            // CSS selector for email
+	PhotoSelector    string            // CSS selector for photo
+	Extractors       []string          // List of extractor types to use
+}
+
+// NewUniversityPatterns creates a new pattern registry with known patterns
+func NewUniversityPatterns() *UniversityPatterns {
+	p := &UniversityPatterns{
+		patterns: make(map[string]UniversityConfig),
+	}
+
+	// Register known university patterns
+	p.registerKnownPatterns()
+
+	return p
+}
+
+// GetConfig returns the configuration for a university domain
+func (p *UniversityPatterns) GetConfig(domain string) *UniversityConfig {
+	// Normalize domain
+	domain = strings.ToLower(domain)
+	domain = strings.TrimPrefix(domain, "www.")
+
+	if config, ok := p.patterns[domain]; ok {
+		return &config
+	}
+
+	// Try partial match
+	for key, config := range p.patterns {
+		if strings.Contains(domain, key) || strings.Contains(key, domain) {
+			return &config
+		}
+	}
+
+	return nil
+}
+
+// registerKnownPatterns registers patterns for known German universities
+func (p *UniversityPatterns) registerKnownPatterns() {
+	// KIT - Karlsruher Institut für Technologie
+	p.patterns["kit.edu"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.kit.edu/kit/fakultaeten.php",
+		},
+		StaffLinkPattern: regexp.MustCompile(`/personen/\d+`),
+		NameSelector:     ".person-name, h1.title",
+		PositionSelector: ".person-position, .position",
+		EmailSelector:    "a[href^='mailto:']",
+		PhotoSelector:    ".person-image img, .portrait img",
+	}
+
+	// TUM - Technische Universität München
+	p.patterns["tum.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.tum.de/die-tum/fakultaeten",
+		},
+		StaffLinkPattern: regexp.MustCompile(`/person/\w+`),
+		NameSelector:     ".person-name, h1",
+		PositionSelector: ".person-title, .function",
+		EmailSelector:    "a[href^='mailto:']",
+		PhotoSelector:    ".person-photo img",
+	}
+
+	// LMU - Ludwig-Maximilians-Universität München
+	p.patterns["lmu.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.lmu.de/de/die-lmu/struktur/fakultaeten-einrichtungen-zentren-und-weitere-institutionen/",
+		},
+		NameSelector:     ".person h2, .staff-name",
+		PositionSelector: ".person-position, .staff-position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// RWTH Aachen
+	p.patterns["rwth-aachen.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.rwth-aachen.de/cms/root/Die-RWTH/Fakultaeten/~ep/Fakultaeten-und-Einrichtungen/",
+		},
+		NameSelector:     ".person-name, h3.title",
+		PositionSelector: ".person-function, .position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// TU Berlin
+	p.patterns["tu-berlin.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.tu.berlin/ueber-die-tu-berlin/organisation/fakultaeten-und-einrichtungen",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".position, .function",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// FU Berlin
+	p.patterns["fu-berlin.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.fu-berlin.de/einrichtungen/fachbereiche/",
+		},
+		NameSelector:     ".person-fullname, h2",
+		PositionSelector: ".person-position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// HU Berlin
+	p.patterns["hu-berlin.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.hu-berlin.de/de/einrichtungen-organisation/fakultaeten-und-institute",
+		},
+		NameSelector:     ".person h2, .name",
+		PositionSelector: ".function, .position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Freiburg
+	p.patterns["uni-freiburg.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://uni-freiburg.de/universitaet/fakultaeten/",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".person-position, .function",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Heidelberg
+	p.patterns["uni-heidelberg.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.uni-heidelberg.de/de/fakultaeten",
+		},
+		NameSelector:     ".person-fullname, h2",
+		PositionSelector: ".person-position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// TU Dresden
+	p.patterns["tu-dresden.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://tu-dresden.de/tu-dresden/organisation/bereiche-und-fakultaeten",
+		},
+		NameSelector:     ".person-name, h2.name",
+		PositionSelector: ".person-function, .funktion",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Leipzig
+	p.patterns["uni-leipzig.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.uni-leipzig.de/universitaet/struktur/fakultaeten",
+		},
+		NameSelector:     ".person h2, .name",
+		PositionSelector: ".position, .funktion",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Köln
+	p.patterns["uni-koeln.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.uni-koeln.de/",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".person-position, .function",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Bonn
+	p.patterns["uni-bonn.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.uni-bonn.de/de/universitaet/fakultaeten",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".person-position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Münster
+	p.patterns["uni-muenster.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.uni-muenster.de/de/fakultaeten.html",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".person-function",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Hamburg
+	p.patterns["uni-hamburg.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.uni-hamburg.de/einrichtungen/fakultaeten.html",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// Universität Göttingen
+	p.patterns["uni-goettingen.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.uni-goettingen.de/de/fakultaeten/27952.html",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".person-position",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+
+	// TU Darmstadt
+	p.patterns["tu-darmstadt.de"] = UniversityConfig{
+		StaffListURLs: []string{
+			"https://www.tu-darmstadt.de/universitaet/fachbereiche/index.de.jsp",
+		},
+		NameSelector:     ".person-name, h2",
+		PositionSelector: ".person-position, .funktion",
+		EmailSelector:    "a[href^='mailto:']",
+	}
+}
+
+// CommonStaffPagePaths returns common paths where staff listings are found
+func CommonStaffPagePaths() []string {
+	return []string{
+		"/personen",
+		"/team",
+		"/mitarbeiter",
+		"/mitarbeitende",
+		"/staff",
+		"/people",
+		"/ueber-uns/team",
+		"/about/team",
+		"/fakultaet/personen",
+		"/institut/mitarbeiter",
+		"/lehrstuhl/team",
+		"/personal",
+		"/beschaeftigte",
+		"/dozenten",
+		"/professoren",
+	}
+}
+
+// CommonPersonSelectors returns common CSS selectors for person elements
+func CommonPersonSelectors() []string {
+	return []string{
+		".person",
+		".person-card",
+		".staff-member",
+		".team-member",
+		".mitarbeiter",
+		".employee",
+		".vcard",
+		".h-card",
+		"[itemtype='http://schema.org/Person']",
+		".person-entry",
+		".staff-entry",
+		".profile-card",
+	}
+}
+
+// TitlePrefixes returns common German academic title prefixes
+func TitlePrefixes() []string {
+	return []string{
+		"Prof. Dr. Dr. h.c. mult.",
+		"Prof. Dr. Dr. h.c.",
+		"Prof. Dr. Dr.",
+		"Prof. Dr.-Ing.",
+		"Prof. Dr. rer. nat.",
+		"Prof. Dr. phil.",
+		"Prof. Dr. jur.",
+		"Prof. Dr. med.",
+		"Prof. Dr.",
+		"Prof.",
+		"PD Dr.",
+		"apl. Prof. Dr.",
+		"Jun.-Prof. Dr.",
+		"Dr.-Ing.",
+		"Dr. rer. nat.",
+		"Dr. phil.",
+		"Dr. jur.",
+		"Dr. med.",
+		"Dr.",
+		"Dipl.-Ing.",
+		"Dipl.-Inf.",
+		"Dipl.-Phys.",
+		"Dipl.-Math.",
+		"Dipl.-Kfm.",
+		"M.Sc.",
+		"M.A.",
+		"M.Eng.",
+		"B.Sc.",
+		"B.A.",
+	}
+}
+
+// PositionKeywords returns keywords that indicate staff positions
+func PositionKeywords() []string {
+	return []string{
+		// Professors
+		"Professor", "Professorin",
+		"Ordinarius",
+		"Lehrstuhlinhaber", "Lehrstuhlinhaberin",
+		"Dekan", "Dekanin",
+		"Rektor", "Rektorin",
+
+		// Research staff
+		"Wissenschaftlicher Mitarbeiter", "Wissenschaftliche Mitarbeiterin",
+		"Akademischer Rat", "Akademische Rätin",
+		"Postdoktorand", "Postdoktorandin",
+		"Doktorand", "Doktorandin",
+		"Promovend", "Promovendin",
+		"Forscher", "Forscherin",
+		"Researcher",
+
+		// Teaching
+		"Dozent", "Dozentin",
+		"Lektor", "Lektorin",
+		"Lehrbeauftragter", "Lehrbeauftragte",
+
+		// Administrative
+		"Sekretär", "Sekretärin",
+		"Geschäftsführer", "Geschäftsführerin",
+		"Verwaltungsleiter", "Verwaltungsleiterin",
+		"Referent", "Referentin",
+
+		// Students
+		"Studentische Hilfskraft",
+		"Wissenschaftliche Hilfskraft",
+		"Tutor", "Tutorin",
+	}
+}
--- a/edu-search-service/internal/staff/publication_adapter.go
+++ b/edu-search-service/internal/staff/publication_adapter.go
@@ -0,0 +1,78 @@
+// Package staff provides university staff and publication crawling functionality
+package staff
+
+import (
+	"context"
+	"log"
+	"time"
+
+	"github.com/google/uuid"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+	"github.com/breakpilot/edu-search-service/internal/orchestrator"
+)
+
+// PublicationOrchestratorAdapter adapts publication crawling to the orchestrator interface
+// Note: This is a stub for now - publication crawling is a future feature
+type PublicationOrchestratorAdapter struct {
+	repo *database.Repository
+}
+
+// NewPublicationOrchestratorAdapter creates a new publication crawler adapter
+func NewPublicationOrchestratorAdapter(repo *database.Repository) *PublicationOrchestratorAdapter {
+	return &PublicationOrchestratorAdapter{
+		repo: repo,
+	}
+}
+
+// CrawlPublicationsForUniversity crawls publications for all staff at a university
+// This is Phase 4: Publication discovery (future implementation)
+func (a *PublicationOrchestratorAdapter) CrawlPublicationsForUniversity(ctx context.Context, universityID uuid.UUID) (*orchestrator.CrawlProgress, error) {
+	start := time.Now()
+	progress := &orchestrator.CrawlProgress{
+		Phase:     orchestrator.PhasePublications,
+		StartedAt: start,
+	}
+
+	log.Printf("[PublicationAdapter] Publications phase for university %s", universityID)
+
+	// Get staff members for this university
+	staffList, err := a.repo.SearchStaff(ctx, database.StaffSearchParams{
+		UniversityID: &universityID,
+		Limit:        10000,
+	})
+	if err != nil {
+		progress.Errors = append(progress.Errors, err.Error())
+		return progress, err
+	}
+
+	log.Printf("[PublicationAdapter] Found %d staff members for publication crawling", staffList.Total)
+
+	// TODO: Implement actual publication crawling
+	// - For each staff member with ORCID/Google Scholar ID:
+	//   - Fetch publications from ORCID API
+	//   - Fetch publications from Google Scholar
+	//   - Match and deduplicate
+	//   - Store in database
+	//
+	// For now, we mark this phase as complete (no-op)
+
+	pubCount := 0
+
+	// Count existing publications for this university
+	for _, staff := range staffList.Staff {
+		pubs, err := a.repo.GetStaffPublications(ctx, staff.ID)
+		if err == nil {
+			pubCount += len(pubs)
+		}
+	}
+
+	progress.ItemsFound = pubCount
+	progress.ItemsProcessed = staffList.Total
+	now := time.Now()
+	progress.CompletedAt = &now
+
+	log.Printf("[PublicationAdapter] Publications phase completed for university %s: %d existing publications found", universityID, pubCount)
+
+	return progress, nil
+}
--- a/edu-search-service/internal/staff/staff_crawler.go
+++ b/edu-search-service/internal/staff/staff_crawler.go
--- a/edu-search-service/internal/staff/staff_crawler_test.go
+++ b/edu-search-service/internal/staff/staff_crawler_test.go
@@ -0,0 +1,348 @@
+package staff
+
+import (
+	"testing"
+
+	"github.com/breakpilot/edu-search-service/internal/database"
+)
+
+func TestParseName_FullName_WithTitle(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name           string
+		fullName       string
+		expectedFirst  string
+		expectedLast   string
+		expectedTitle  bool
+	}{
+		{
+			name:           "Prof. Dr. with first and last name",
+			fullName:       "Prof. Dr. Hans Müller",
+			expectedFirst:  "Hans",
+			expectedLast:   "Müller",
+			expectedTitle:  true,
+		},
+		{
+			name:           "Dr. with first and last name",
+			fullName:       "Dr. Maria Schmidt",
+			expectedFirst:  "Maria",
+			expectedLast:   "Schmidt",
+			expectedTitle:  true,
+		},
+		{
+			name:           "Simple name without title",
+			fullName:       "Thomas Weber",
+			expectedFirst:  "Thomas",
+			expectedLast:   "Weber",
+			expectedTitle:  false,
+		},
+		{
+			name:           "Multiple first names",
+			fullName:       "Prof. Dr. Hans-Peter Meier",
+			expectedFirst:  "Hans-Peter",
+			expectedLast:   "Meier",
+			expectedTitle:  true,
+		},
+		{
+			name:           "Single name",
+			fullName:       "Müller",
+			expectedFirst:  "",
+			expectedLast:   "Müller",
+			expectedTitle:  false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			person := &database.UniversityStaff{}
+			crawler.parseName(tt.fullName, person)
+
+			firstName := ""
+			if person.FirstName != nil {
+				firstName = *person.FirstName
+			}
+
+			if firstName != tt.expectedFirst {
+				t.Errorf("First name: expected %q, got %q", tt.expectedFirst, firstName)
+			}
+			if person.LastName != tt.expectedLast {
+				t.Errorf("Last name: expected %q, got %q", tt.expectedLast, person.LastName)
+			}
+			hasTitle := person.Title != nil && *person.Title != ""
+			if hasTitle != tt.expectedTitle {
+				t.Errorf("Has title: expected %v, got %v", tt.expectedTitle, hasTitle)
+			}
+		})
+	}
+}
+
+func TestClassifyPosition_Professor(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+		expected string
+	}{
+		{"Full Professor", "Professor für Informatik", "professor"},
+		{"Prof abbreviation", "Prof. Dr. Müller", "professor"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.classifyPosition(tt.position)
+			if result == nil {
+				t.Errorf("Expected %q, got nil for position %q", tt.expected, tt.position)
+				return
+			}
+			if *result != tt.expected {
+				t.Errorf("Expected %q, got %q for position %q", tt.expected, *result, tt.position)
+			}
+		})
+	}
+}
+
+func TestClassifyPosition_Postdoc(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+		expected string
+	}{
+		{"Postdoc", "Postdoc in Machine Learning", "postdoc"},
+		{"Post-Doc hyphenated", "Post-Doc", "postdoc"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.classifyPosition(tt.position)
+			if result == nil {
+				t.Errorf("Expected %q, got nil for position %q", tt.expected, tt.position)
+				return
+			}
+			if *result != tt.expected {
+				t.Errorf("Expected %q, got %q for position %q", tt.expected, *result, tt.position)
+			}
+		})
+	}
+}
+
+func TestClassifyPosition_PhDStudent(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+		expected string
+	}{
+		{"Doktorand", "Doktorand", "phd_student"},
+		{"PhD Student", "PhD Student", "phd_student"},
+		{"Promovend", "Promovend", "phd_student"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.classifyPosition(tt.position)
+			if result == nil {
+				t.Errorf("Expected %q, got nil for position %q", tt.expected, tt.position)
+				return
+			}
+			if *result != tt.expected {
+				t.Errorf("Expected %q, got %q for position %q", tt.expected, *result, tt.position)
+			}
+		})
+	}
+}
+
+func TestClassifyPosition_Admin(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+		expected string
+	}{
+		{"Sekretariat", "Sekretärin", "admin"},
+		{"Verwaltung", "Verwaltung", "admin"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.classifyPosition(tt.position)
+			if result == nil {
+				t.Errorf("Expected %q, got nil for position %q", tt.expected, tt.position)
+				return
+			}
+			if *result != tt.expected {
+				t.Errorf("Expected %q, got %q for position %q", tt.expected, *result, tt.position)
+			}
+		})
+	}
+}
+
+func TestClassifyPosition_Researcher(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+		expected string
+	}{
+		{"Wissenschaftlicher Mitarbeiter", "Wissenschaftlicher Mitarbeiter", "researcher"},
+		{"Researcher", "Senior Researcher", "researcher"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.classifyPosition(tt.position)
+			if result == nil {
+				t.Errorf("Expected %q, got nil for position %q", tt.expected, tt.position)
+				return
+			}
+			if *result != tt.expected {
+				t.Errorf("Expected %q, got %q for position %q", tt.expected, *result, tt.position)
+			}
+		})
+	}
+}
+
+func TestClassifyPosition_Student(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+		expected string
+	}{
+		{"Studentische Hilfskraft", "Studentische Hilfskraft", "student"},
+		{"HiWi", "Student (HiWi)", "student"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.classifyPosition(tt.position)
+			if result == nil {
+				t.Errorf("Expected %q, got nil for position %q", tt.expected, tt.position)
+				return
+			}
+			if *result != tt.expected {
+				t.Errorf("Expected %q, got %q for position %q", tt.expected, *result, tt.position)
+			}
+		})
+	}
+}
+
+func TestIsProfessor_True(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+	}{
+		{"Professor keyword", "Professor für Mathematik"},
+		{"Prof. abbreviation", "Prof. Dr. Müller"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.isProfessor(tt.position)
+			if !result {
+				t.Errorf("Expected true for position=%q", tt.position)
+			}
+		})
+	}
+}
+
+func TestIsProfessor_False(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name     string
+		position string
+	}{
+		{"Dr. only", "Dr. Wissenschaftlicher Mitarbeiter"},
+		{"Doktorand", "Doktorand"},
+		{"Technical staff", "Laboringenieur"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.isProfessor(tt.position)
+			if result {
+				t.Errorf("Expected false for position=%q", tt.position)
+			}
+		})
+	}
+}
+
+func TestLooksLikePosition_True(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name string
+		text string
+	}{
+		{"Professor", "Professor für Informatik"},
+		{"Wissenschaftlicher Mitarbeiter", "Wissenschaftlicher Mitarbeiter"},
+		{"Doktorand", "Doktorand"},
+		{"Sekretär", "Sekretärin"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.looksLikePosition(tt.text)
+			if !result {
+				t.Errorf("Expected true for text=%q", tt.text)
+			}
+		})
+	}
+}
+
+func TestLooksLikePosition_False(t *testing.T) {
+	crawler := &StaffCrawler{}
+
+	tests := []struct {
+		name string
+		text string
+	}{
+		{"Name", "Hans Müller"},
+		{"Email", "test@example.com"},
+		{"Random text", "Room 123"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := crawler.looksLikePosition(tt.text)
+			if result {
+				t.Errorf("Expected false for text=%q", tt.text)
+			}
+		})
+	}
+}
+
+func TestResolveURL(t *testing.T) {
+	tests := []struct {
+		name     string
+		baseURL  string
+		href     string
+		expected string
+	}{
+		{"Absolute URL", "https://example.com", "https://other.com/page", "https://other.com/page"},
+		{"Relative path", "https://example.com/team", "/person/123", "https://example.com/person/123"},
+		{"Relative no slash", "https://example.com/team/", "member", "https://example.com/team/member"},
+		{"Empty href", "https://example.com", "", ""},
+		{"Root relative", "https://example.com/a/b/c", "/root", "https://example.com/root"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := resolveURL(tt.baseURL, tt.href)
+			if result != tt.expected {
+				t.Errorf("resolveURL(%q, %q) = %q, expected %q",
+					tt.baseURL, tt.href, result, tt.expected)
+			}
+		})
+	}
+}
--- a/edu-search-service/internal/tagger/tagger.go
+++ b/edu-search-service/internal/tagger/tagger.go
@@ -0,0 +1,455 @@
+package tagger
+
+import (
+	"os"
+	"path/filepath"
+	"regexp"
+	"sort"
+	"strings"
+
+	"gopkg.in/yaml.v3"
+)
+
+// TagResult contains all tags assigned to a document
+type TagResult struct {
+	DocType    string   `json:"doc_type"`
+	Subjects   []string `json:"subjects"`
+	SchoolLevel string  `json:"school_level"`
+	State      string   `json:"state"`
+	TrustScore float64  `json:"trust_score"`
+}
+
+// Tagger applies rules to content and URLs
+type Tagger struct {
+	docTypeRules  *DocTypeRules
+	subjectRules  *SubjectRules
+	levelRules    *LevelRules
+	trustRules    *TrustRules
+}
+
+// DocTypeRules YAML structure
+type DocTypeRules struct {
+	DocTypes      map[string]DocTypeRule `yaml:"doc_types"`
+	PriorityOrder []string               `yaml:"priority_order"`
+}
+
+type DocTypeRule struct {
+	StrongTerms []string `yaml:"strong_terms"`
+	MediumTerms []string `yaml:"medium_terms"`
+	URLPatterns []string `yaml:"url_patterns"`
+}
+
+// SubjectRules YAML structure
+type SubjectRules struct {
+	Subjects    map[string]SubjectRule `yaml:"subjects"`
+	Threshold   int                    `yaml:"threshold"`
+	MaxSubjects int                    `yaml:"max_subjects"`
+}
+
+type SubjectRule struct {
+	Strong   []string `yaml:"strong"`
+	Weak     []string `yaml:"weak"`
+	Negative []string `yaml:"negative"`
+}
+
+// LevelRules YAML structure
+type LevelRules struct {
+	Levels        map[string]LevelRule `yaml:"levels"`
+	PriorityOrder []string             `yaml:"priority_order"`
+}
+
+type LevelRule struct {
+	Strong   []string `yaml:"strong"`
+	Weak     []string `yaml:"weak"`
+	Negative []string `yaml:"negative"`
+}
+
+// TrustRules YAML structure
+type TrustRules struct {
+	DomainBoosts     []DomainBoost     `yaml:"domain_boosts"`
+	TLDBoosts        []TLDBoost        `yaml:"tld_boosts"`
+	Penalties        []Penalty         `yaml:"penalties"`
+	ContentPenalties []ContentPenalty  `yaml:"content_penalties"`
+}
+
+type DomainBoost struct {
+	Match  string  `yaml:"match"`
+	Add    float64 `yaml:"add"`
+	Reason string  `yaml:"reason"`
+}
+
+type TLDBoost struct {
+	TLD    string  `yaml:"tld"`
+	Add    float64 `yaml:"add"`
+	Reason string  `yaml:"reason"`
+}
+
+type Penalty struct {
+	IfURLContains []string `yaml:"if_url_contains"`
+	Add           float64  `yaml:"add"`
+	Reason        string   `yaml:"reason"`
+}
+
+type ContentPenalty struct {
+	IfAdDensityGT     *float64 `yaml:"if_ad_density_gt,omitempty"`
+	IfLinkDensityGT   *float64 `yaml:"if_link_density_gt,omitempty"`
+	IfContentLengthLT *int     `yaml:"if_content_length_lt,omitempty"`
+	Add               float64  `yaml:"add"`
+	Reason            string   `yaml:"reason"`
+}
+
+// ContentFeatures for trust scoring
+type ContentFeatures struct {
+	AdDensity     float64
+	LinkDensity   float64
+	ContentLength int
+}
+
+// NewTagger creates a new tagger with rules from the specified directory
+func NewTagger(rulesDir string) (*Tagger, error) {
+	t := &Tagger{}
+
+	// Load doc type rules
+	docTypeBytes, err := os.ReadFile(filepath.Join(rulesDir, "doc_type_rules.yaml"))
+	if err != nil {
+		return nil, err
+	}
+	t.docTypeRules = &DocTypeRules{}
+	if err := yaml.Unmarshal(docTypeBytes, t.docTypeRules); err != nil {
+		return nil, err
+	}
+
+	// Load subject rules
+	subjectBytes, err := os.ReadFile(filepath.Join(rulesDir, "subject_rules.yaml"))
+	if err != nil {
+		return nil, err
+	}
+	t.subjectRules = &SubjectRules{}
+	if err := yaml.Unmarshal(subjectBytes, t.subjectRules); err != nil {
+		return nil, err
+	}
+
+	// Load level rules
+	levelBytes, err := os.ReadFile(filepath.Join(rulesDir, "level_rules.yaml"))
+	if err != nil {
+		return nil, err
+	}
+	t.levelRules = &LevelRules{}
+	if err := yaml.Unmarshal(levelBytes, t.levelRules); err != nil {
+		return nil, err
+	}
+
+	// Load trust rules
+	trustBytes, err := os.ReadFile(filepath.Join(rulesDir, "trust_rules.yaml"))
+	if err != nil {
+		return nil, err
+	}
+	t.trustRules = &TrustRules{}
+	if err := yaml.Unmarshal(trustBytes, t.trustRules); err != nil {
+		return nil, err
+	}
+
+	return t, nil
+}
+
+// Tag applies all rules to content and returns tags
+func (t *Tagger) Tag(url string, title string, content string, features ContentFeatures) TagResult {
+	lowerURL := strings.ToLower(url)
+	lowerTitle := strings.ToLower(title)
+	lowerContent := strings.ToLower(content)
+	combined := lowerTitle + " " + lowerContent
+
+	result := TagResult{
+		DocType:     "Sonstiges",
+		Subjects:    []string{},
+		SchoolLevel: "NA",
+		State:       t.detectState(lowerURL),
+	}
+
+	// Tag doc type
+	result.DocType = t.tagDocType(lowerURL, combined)
+
+	// Tag subjects
+	result.Subjects = t.tagSubjects(combined)
+
+	// Tag school level
+	result.SchoolLevel = t.tagSchoolLevel(combined)
+
+	// Calculate trust score
+	result.TrustScore = t.calculateTrustScore(lowerURL, features)
+
+	return result
+}
+
+func (t *Tagger) tagDocType(url string, content string) string {
+	scores := make(map[string]int)
+
+	for docType, rule := range t.docTypeRules.DocTypes {
+		score := 0
+
+		// Check strong terms (+4 each)
+		for _, term := range rule.StrongTerms {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score += 4
+			}
+		}
+
+		// Check medium terms (+3 each)
+		for _, term := range rule.MediumTerms {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score += 3
+			}
+		}
+
+		// Check URL patterns (+2 each)
+		for _, pattern := range rule.URLPatterns {
+			if strings.Contains(url, strings.ToLower(pattern)) {
+				score += 2
+			}
+		}
+
+		if score > 0 {
+			scores[docType] = score
+		}
+	}
+
+	if len(scores) == 0 {
+		return "Sonstiges"
+	}
+
+	// Find highest scoring type, respecting priority for ties
+	var bestType string
+	bestScore := 0
+
+	for _, docType := range t.docTypeRules.PriorityOrder {
+		if score, ok := scores[docType]; ok {
+			if score > bestScore || (score == bestScore && bestType == "") {
+				bestScore = score
+				bestType = docType
+			}
+		}
+	}
+
+	if bestType == "" {
+		return "Sonstiges"
+	}
+
+	return bestType
+}
+
+func (t *Tagger) tagSubjects(content string) []string {
+	type subjectScore struct {
+		name  string
+		score int
+	}
+
+	var scores []subjectScore
+
+	for subject, rule := range t.subjectRules.Subjects {
+		score := 0
+
+		// Check strong terms (+3 each)
+		for _, term := range rule.Strong {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score += 3
+			}
+		}
+
+		// Check weak terms (+1 each)
+		for _, term := range rule.Weak {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score += 1
+			}
+		}
+
+		// Check negative terms (-2 each)
+		for _, term := range rule.Negative {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score -= 2
+			}
+		}
+
+		threshold := t.subjectRules.Threshold
+		if threshold == 0 {
+			threshold = 4 // default
+		}
+
+		if score >= threshold {
+			scores = append(scores, subjectScore{name: subject, score: score})
+		}
+	}
+
+	// Sort by score descending
+	sort.Slice(scores, func(i, j int) bool {
+		return scores[i].score > scores[j].score
+	})
+
+	// Take top N subjects
+	maxSubjects := t.subjectRules.MaxSubjects
+	if maxSubjects == 0 {
+		maxSubjects = 3 // default
+	}
+
+	var result []string
+	for i, s := range scores {
+		if i >= maxSubjects {
+			break
+		}
+		result = append(result, s.name)
+	}
+
+	return result
+}
+
+func (t *Tagger) tagSchoolLevel(content string) string {
+	scores := make(map[string]int)
+
+	for level, rule := range t.levelRules.Levels {
+		score := 0
+
+		// Check strong terms (+3 each)
+		for _, term := range rule.Strong {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score += 3
+			}
+		}
+
+		// Check weak terms (+1 each)
+		for _, term := range rule.Weak {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score += 1
+			}
+		}
+
+		// Check negative terms (-2 each)
+		for _, term := range rule.Negative {
+			if strings.Contains(content, strings.ToLower(term)) {
+				score -= 2
+			}
+		}
+
+		if score > 0 {
+			scores[level] = score
+		}
+	}
+
+	if len(scores) == 0 {
+		return "NA"
+	}
+
+	// Find highest scoring level, respecting priority for ties
+	var bestLevel string
+	bestScore := 0
+
+	for _, level := range t.levelRules.PriorityOrder {
+		if score, ok := scores[level]; ok {
+			if score > bestScore {
+				bestScore = score
+				bestLevel = level
+			}
+		}
+	}
+
+	if bestLevel == "" {
+		return "NA"
+	}
+
+	return bestLevel
+}
+
+func (t *Tagger) calculateTrustScore(url string, features ContentFeatures) float64 {
+	score := 0.50 // base score
+
+	// Apply domain boosts
+	for _, boost := range t.trustRules.DomainBoosts {
+		if matchDomainPattern(url, boost.Match) {
+			score += boost.Add
+		}
+	}
+
+	// Apply TLD boosts
+	for _, boost := range t.trustRules.TLDBoosts {
+		if strings.HasSuffix(url, boost.TLD) || strings.Contains(url, boost.TLD+"/") {
+			score += boost.Add
+		}
+	}
+
+	// Apply URL penalties
+	for _, penalty := range t.trustRules.Penalties {
+		for _, pattern := range penalty.IfURLContains {
+			if strings.Contains(url, strings.ToLower(pattern)) {
+				score += penalty.Add // Add is negative
+				break
+			}
+		}
+	}
+
+	// Apply content penalties
+	for _, penalty := range t.trustRules.ContentPenalties {
+		if penalty.IfAdDensityGT != nil && features.AdDensity > *penalty.IfAdDensityGT {
+			score += penalty.Add
+		}
+		if penalty.IfLinkDensityGT != nil && features.LinkDensity > *penalty.IfLinkDensityGT {
+			score += penalty.Add
+		}
+		if penalty.IfContentLengthLT != nil && features.ContentLength < *penalty.IfContentLengthLT {
+			score += penalty.Add
+		}
+	}
+
+	// Clamp to [0, 1]
+	if score < 0 {
+		score = 0
+	}
+	if score > 1 {
+		score = 1
+	}
+
+	return score
+}
+
+func matchDomainPattern(url string, pattern string) bool {
+	// Convert wildcard pattern to regex
+	// *.example.de should match subdomain.example.de and example.de
+	regexPattern := strings.ReplaceAll(pattern, ".", "\\.")
+	regexPattern = strings.ReplaceAll(regexPattern, "*", ".*")
+	regexPattern = "(?i)" + regexPattern // case insensitive
+
+	re, err := regexp.Compile(regexPattern)
+	if err != nil {
+		return false
+	}
+
+	return re.MatchString(url)
+}
+
+func (t *Tagger) detectState(url string) string {
+	statePatterns := map[string][]string{
+		"BW":  {"baden-wuerttemberg", "bw.de", "schule-bw.de", "kultusministerium.baden"},
+		"BY":  {"bayern.de", "isb.bayern", "km.bayern"},
+		"BE":  {"berlin.de", "bildungsserver.berlin"},
+		"BB":  {"brandenburg.de", "bildungsserver.brandenburg"},
+		"HB":  {"bremen.de", "lis.bremen"},
+		"HH":  {"hamburg.de", "li.hamburg"},
+		"HE":  {"hessen.de", "hkm.hessen", "bildung.hessen"},
+		"MV":  {"mecklenburg-vorpommern", "mv.de", "bildung-mv.de"},
+		"NI":  {"niedersachsen.de", "nibis.de", "mk.niedersachsen"},
+		"NW":  {"nrw.de", "learnline.nrw", "schulministerium.nrw"},
+		"RP":  {"rheinland-pfalz", "rlp.de", "bildung-rp.de"},
+		"SL":  {"saarland.de", "bildungsserver.saarland"},
+		"SN":  {"sachsen.de", "schule.sachsen", "smk.sachsen"},
+		"ST":  {"sachsen-anhalt", "bildung-lsa.de", "mk.sachsen-anhalt"},
+		"SH":  {"schleswig-holstein", "sh.de", "bildungsserver.schleswig"},
+		"TH":  {"thueringen.de", "schulportal-thueringen"},
+	}
+
+	for state, patterns := range statePatterns {
+		for _, pattern := range patterns {
+			if strings.Contains(url, pattern) {
+				return state
+			}
+		}
+	}
+
+	return "" // Bundesweit or unknown
+}
--- a/edu-search-service/internal/tagger/tagger_test.go
+++ b/edu-search-service/internal/tagger/tagger_test.go
@@ -0,0 +1,557 @@
+package tagger
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// createTestRulesDir creates temporary test rule files
+func createTestRulesDir(t *testing.T) string {
+	t.Helper()
+
+	dir := t.TempDir()
+
+	// Create doc_type_rules.yaml
+	docTypeRules := `doc_types:
+  Lehrplan:
+    strong_terms:
+      - Lehrplan
+      - Kernlehrplan
+      - Bildungsplan
+    medium_terms:
+      - Curriculum
+    url_patterns:
+      - /lehrplan
+  Arbeitsblatt:
+    strong_terms:
+      - Arbeitsblatt
+      - Übungsblatt
+    medium_terms:
+      - Aufgaben
+    url_patterns:
+      - /arbeitsblatt
+  Studie_Bericht:
+    strong_terms:
+      - Studie
+      - PISA
+    medium_terms:
+      - Ergebnis
+    url_patterns:
+      - /studie
+priority_order:
+  - Lehrplan
+  - Arbeitsblatt
+  - Studie_Bericht
+  - Sonstiges
+`
+	if err := os.WriteFile(filepath.Join(dir, "doc_type_rules.yaml"), []byte(docTypeRules), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	// Create subject_rules.yaml
+	subjectRules := `subjects:
+  Mathematik:
+    strong:
+      - Mathematik
+      - Algebra
+      - Geometrie
+    weak:
+      - rechnen
+      - Zahlen
+    negative:
+      - Geschichte der Mathematik
+  Deutsch:
+    strong:
+      - Deutsch
+      - Grammatik
+      - Rechtschreibung
+    weak:
+      - Lesen
+      - Schreiben
+    negative:
+      - Deutsch als Fremdsprache
+  Geschichte:
+    strong:
+      - Geschichte
+      - Historisch
+    weak:
+      - Epoche
+      - Jahrhundert
+    negative:
+      - Naturgeschichte
+threshold: 4
+max_subjects: 3
+`
+	if err := os.WriteFile(filepath.Join(dir, "subject_rules.yaml"), []byte(subjectRules), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	// Create level_rules.yaml
+	levelRules := `levels:
+  Grundschule:
+    strong:
+      - Grundschule
+      - Primarstufe
+      - Klasse 1-4
+    weak:
+      - Erstklässler
+    negative:
+      - Sekundarstufe
+  Gymnasium:
+    strong:
+      - Gymnasium
+      - Abitur
+      - Oberstufe
+    weak:
+      - Sekundarstufe II
+    negative:
+      - Realschule
+  Sek_I:
+    strong:
+      - Sekundarstufe I
+      - Klasse 5-10
+      - Hauptschule
+    weak:
+      - Mittelstufe
+    negative:
+      - Grundschule
+priority_order:
+  - Gymnasium
+  - Sek_I
+  - Grundschule
+  - NA
+`
+	if err := os.WriteFile(filepath.Join(dir, "level_rules.yaml"), []byte(levelRules), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	// Create trust_rules.yaml
+	trustRules := `domain_boosts:
+  - match: "*.kmk.org"
+    add: 0.30
+    reason: "Kultusministerkonferenz"
+  - match: "*.bildungsserver.de"
+    add: 0.25
+    reason: "Deutscher Bildungsserver"
+  - match: "*.bayern.de"
+    add: 0.20
+    reason: "Bayerische Landesregierung"
+tld_boosts:
+  - tld: ".gov"
+    add: 0.15
+    reason: "Government domain"
+penalties:
+  - if_url_contains:
+      - "forum"
+      - "blog"
+    add: -0.10
+    reason: "User generated content"
+content_penalties:
+  - if_ad_density_gt: 0.3
+    add: -0.15
+    reason: "High ad density"
+  - if_content_length_lt: 200
+    add: -0.10
+    reason: "Very short content"
+`
+	if err := os.WriteFile(filepath.Join(dir, "trust_rules.yaml"), []byte(trustRules), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	return dir
+}
+
+func TestNewTagger_Success(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatalf("NewTagger failed: %v", err)
+	}
+
+	if tagger == nil {
+		t.Fatal("Expected non-nil tagger")
+	}
+
+	if tagger.docTypeRules == nil {
+		t.Error("docTypeRules not loaded")
+	}
+	if tagger.subjectRules == nil {
+		t.Error("subjectRules not loaded")
+	}
+	if tagger.levelRules == nil {
+		t.Error("levelRules not loaded")
+	}
+	if tagger.trustRules == nil {
+		t.Error("trustRules not loaded")
+	}
+}
+
+func TestNewTagger_MissingFile(t *testing.T) {
+	_, err := NewTagger("/nonexistent/path")
+	if err == nil {
+		t.Error("Expected error for nonexistent rules directory")
+	}
+}
+
+func TestTagger_TagDocType_Lehrplan(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	tests := []struct {
+		name     string
+		url      string
+		content  string
+		expected string
+	}{
+		{
+			name:     "Strong term in content",
+			url:      "https://example.com/page",
+			content:  "Dies ist der Lehrplan für Mathematik in der Sekundarstufe",
+			expected: "Lehrplan",
+		},
+		{
+			name:     "URL pattern match",
+			url:      "https://example.com/lehrplan/mathe",
+			content:  "Allgemeine Informationen zum Fach",
+			expected: "Lehrplan",
+		},
+		{
+			name:     "Multiple strong terms",
+			url:      "https://example.com/bildung",
+			content:  "Kernlehrplan und Bildungsplan für das Curriculum",
+			expected: "Lehrplan",
+		},
+		{
+			name:     "Arbeitsblatt detection",
+			url:      "https://example.com/material",
+			content:  "Arbeitsblatt zum Thema Rechnen mit Übungsblatt",
+			expected: "Arbeitsblatt",
+		},
+		{
+			name:     "No match returns Sonstiges",
+			url:      "https://example.com/page",
+			content:  "Eine allgemeine Webseite ohne spezifische Bildungsinhalte",
+			expected: "Sonstiges",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := tagger.Tag(tt.url, "", tt.content, ContentFeatures{})
+			if result.DocType != tt.expected {
+				t.Errorf("Expected DocType %q, got %q", tt.expected, result.DocType)
+			}
+		})
+	}
+}
+
+func TestTagger_TagSubjects(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	tests := []struct {
+		name            string
+		content         string
+		expectedContain []string
+		expectedMissing []string
+	}{
+		{
+			name:            "Mathematik detection",
+			content:         "In Mathematik lernen wir Algebra und Geometrie sowie das Rechnen mit Zahlen",
+			expectedContain: []string{"Mathematik"},
+		},
+		{
+			name:            "Deutsch detection",
+			content:         "Im Fach Deutsch geht es um Grammatik, Rechtschreibung und das Lesen von Texten",
+			expectedContain: []string{"Deutsch"},
+		},
+		{
+			name:            "Multiple subjects",
+			content:         "Mathematik und Algebra verbinden sich mit Geschichte und historischen Epochen",
+			expectedContain: []string{"Mathematik", "Geschichte"},
+		},
+		{
+			name:            "No subjects detected",
+			content:         "Ein Text ohne spezifische Fachbegriffe",
+			expectedContain: []string{},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := tagger.Tag("https://example.com", "", tt.content, ContentFeatures{})
+
+			for _, expected := range tt.expectedContain {
+				found := false
+				for _, subject := range result.Subjects {
+					if subject == expected {
+						found = true
+						break
+					}
+				}
+				if !found {
+					t.Errorf("Expected subject %q not found in %v", expected, result.Subjects)
+				}
+			}
+		})
+	}
+}
+
+func TestTagger_TagSchoolLevel(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	tests := []struct {
+		name     string
+		content  string
+		expected string
+	}{
+		{
+			name:     "Grundschule detection",
+			content:  "Material für die Grundschule und Primarstufe",
+			expected: "Grundschule",
+		},
+		{
+			name:     "Gymnasium detection",
+			content:  "Vorbereitung auf das Abitur am Gymnasium in der Oberstufe",
+			expected: "Gymnasium",
+		},
+		{
+			name:     "Sekundarstufe I detection",
+			content:  "Aufgaben für Sekundarstufe I in Klasse 5-10",
+			expected: "Sek_I",
+		},
+		{
+			name:     "No level detected",
+			content:  "Allgemeine Bildungsinformationen",
+			expected: "NA",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := tagger.Tag("https://example.com", "", tt.content, ContentFeatures{})
+			if result.SchoolLevel != tt.expected {
+				t.Errorf("Expected SchoolLevel %q, got %q", tt.expected, result.SchoolLevel)
+			}
+		})
+	}
+}
+
+func TestTagger_TrustScore(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	tests := []struct {
+		name        string
+		url         string
+		features    ContentFeatures
+		minExpected float64
+		maxExpected float64
+	}{
+		{
+			name:        "Base score for unknown domain",
+			url:         "https://unknown-domain.com/page",
+			features:    ContentFeatures{ContentLength: 500},
+			minExpected: 0.40,
+			maxExpected: 0.60,
+		},
+		{
+			name:        "KMK domain boost",
+			url:         "https://www.kmk.org/bildung",
+			features:    ContentFeatures{ContentLength: 500},
+			minExpected: 0.70,
+			maxExpected: 0.90,
+		},
+		{
+			name:        "Bayern domain boost",
+			url:         "https://www.km.bayern.de/lehrplan",
+			features:    ContentFeatures{ContentLength: 500},
+			minExpected: 0.60,
+			maxExpected: 0.80,
+		},
+		{
+			name:        "Forum penalty",
+			url:         "https://example.com/forum/thread",
+			features:    ContentFeatures{ContentLength: 500},
+			minExpected: 0.30,
+			maxExpected: 0.50,
+		},
+		{
+			name:        "High ad density penalty",
+			url:         "https://example.com/page",
+			features:    ContentFeatures{AdDensity: 0.5, ContentLength: 500},
+			minExpected: 0.25,
+			maxExpected: 0.50,
+		},
+		{
+			name:        "Short content penalty",
+			url:         "https://example.com/page",
+			features:    ContentFeatures{ContentLength: 100},
+			minExpected: 0.30,
+			maxExpected: 0.50,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := tagger.Tag(tt.url, "", "Some content text", tt.features)
+			if result.TrustScore < tt.minExpected || result.TrustScore > tt.maxExpected {
+				t.Errorf("TrustScore %f not in expected range [%f, %f]",
+					result.TrustScore, tt.minExpected, tt.maxExpected)
+			}
+		})
+	}
+}
+
+func TestTagger_DetectState(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	tests := []struct {
+		name     string
+		url      string
+		expected string
+	}{
+		{
+			name:     "Bayern detection",
+			url:      "https://www.km.bayern.de/lehrplan",
+			expected: "BY",
+		},
+		{
+			name:     "NRW detection",
+			url:      "https://www.schulministerium.nrw.de/themen",
+			expected: "NW",
+		},
+		{
+			name:     "Berlin detection",
+			url:      "https://www.berlin.de/sen/bildung/schule",
+			expected: "BE",
+		},
+		{
+			name:     "Hessen detection",
+			url:      "https://kultusministerium.hessen.de",
+			expected: "HE",
+		},
+		{
+			name:     "No state (federal)",
+			url:      "https://www.kmk.org/bildung",
+			expected: "",
+		},
+		{
+			name:     "Unknown domain",
+			url:      "https://www.example.com/page",
+			expected: "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := tagger.Tag(tt.url, "", "Some content", ContentFeatures{})
+			if result.State != tt.expected {
+				t.Errorf("Expected State %q, got %q", tt.expected, result.State)
+			}
+		})
+	}
+}
+
+func TestMatchDomainPattern(t *testing.T) {
+	tests := []struct {
+		name     string
+		url      string
+		pattern  string
+		expected bool
+	}{
+		{
+			name:     "Exact match",
+			url:      "https://kmk.org/page",
+			pattern:  "kmk.org",
+			expected: true,
+		},
+		{
+			name:     "Wildcard subdomain",
+			url:      "https://www.kmk.org/page",
+			pattern:  "*.kmk.org",
+			expected: true,
+		},
+		{
+			name:     "No match",
+			url:      "https://example.com/page",
+			pattern:  "*.kmk.org",
+			expected: false,
+		},
+		{
+			name:     "Case insensitive",
+			url:      "https://WWW.KMK.ORG/page",
+			pattern:  "*.kmk.org",
+			expected: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := matchDomainPattern(tt.url, tt.pattern)
+			if result != tt.expected {
+				t.Errorf("matchDomainPattern(%q, %q) = %v, expected %v",
+					tt.url, tt.pattern, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestTagger_CombinedTitleAndContent(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Test that title is combined with content for tagging
+	result := tagger.Tag(
+		"https://example.com/page",
+		"Lehrplan Mathematik Bayern", // Title with keywords
+		"Allgemeiner Text ohne spezifische Begriffe", // Content without keywords
+		ContentFeatures{ContentLength: 500},
+	)
+
+	if result.DocType != "Lehrplan" {
+		t.Errorf("Expected DocType 'Lehrplan' from title, got %q", result.DocType)
+	}
+}
+
+func TestTrustScoreClamping(t *testing.T) {
+	rulesDir := createTestRulesDir(t)
+	tagger, err := NewTagger(rulesDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Test that score is clamped to [0, 1]
+	result := tagger.Tag(
+		"https://www.kmk.org/page", // High trust domain
+		"",
+		"Content",
+		ContentFeatures{ContentLength: 1000},
+	)
+
+	if result.TrustScore < 0 || result.TrustScore > 1 {
+		t.Errorf("TrustScore %f should be in range [0, 1]", result.TrustScore)
+	}
+}
--- a/edu-search-service/policies/bundeslaender.yaml
+++ b/edu-search-service/policies/bundeslaender.yaml
@@ -0,0 +1,347 @@
+# =============================================================================
+# Source-Policy System - Initial Data Configuration
+# =============================================================================
+# This file contains the initial whitelist of allowed data sources for the
+# edu-search-service. All sources must be official Open-Data portals or
+# government sources under §5 UrhG (German Copyright Act).
+#
+# IMPORTANT:
+# - Training with external data is FORBIDDEN (training: allowed: false)
+# - All changes are logged in the audit trail
+# - PII is blocked automatically
+# =============================================================================
+
+# =============================================================================
+# FEDERAL / KMK (Bundesebene)
+# =============================================================================
+federal:
+  name: "KMK & Bundesebene"
+  sources:
+    # Kultusministerkonferenz
+    - domain: "kmk.org"
+      name: "Kultusministerkonferenz"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: KMK, {title}, {date}"
+      trust_boost: 0.95
+
+    # Deutscher Bildungsserver
+    - domain: "bildungsserver.de"
+      name: "Deutscher Bildungsserver"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: Deutscher Bildungsserver, {title}, {date}"
+      trust_boost: 0.90
+
+    # IQB (Institut zur Qualitaetsentwicklung im Bildungswesen)
+    - domain: "iqb.hu-berlin.de"
+      name: "IQB Bildungstrends"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: IQB, {title}, {date}"
+      trust_boost: 0.90
+
+    # BMBF (Bundesministerium fuer Bildung und Forschung)
+    - domain: "bmbf.de"
+      name: "Bundesministerium fuer Bildung und Forschung"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: BMBF, {title}, {date}"
+      trust_boost: 0.95
+
+# =============================================================================
+# NIEDERSACHSEN (NI)
+# =============================================================================
+NI:
+  name: "Niedersachsen"
+  sources:
+    - domain: "nibis.de"
+      name: "NiBiS Bildungsserver"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: NiBiS, {title}, {date}"
+      trust_boost: 0.85
+
+    - domain: "mk.niedersachsen.de"
+      name: "Kultusministerium Niedersachsen"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: MK Niedersachsen, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "cuvo.nibis.de"
+      name: "Kerncurricula Niedersachsen"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: Kerncurriculum Niedersachsen, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "nline.nibis.de"
+      name: "NiBiS Online-Materialien"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: NiBiS, {title}, {date}"
+      trust_boost: 0.80
+
+# =============================================================================
+# BAYERN (BY)
+# =============================================================================
+BY:
+  name: "Bayern"
+  sources:
+    - domain: "km.bayern.de"
+      name: "Bayerisches Kultusministerium"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: StMUK Bayern, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "isb.bayern.de"
+      name: "ISB Bayern"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: ISB Bayern, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "lehrplanplus.bayern.de"
+      name: "LehrplanPLUS"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: LehrplanPLUS Bayern, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "mebis.bayern.de"
+      name: "mebis Landesmedienzentrum"
+      license: "CC-BY-SA"
+      legal_basis: "Creative Commons"
+      citation_template: "Quelle: mebis Bayern, {title}, {date}"
+      trust_boost: 0.75
+
+# =============================================================================
+# BADEN-WUERTTEMBERG (BW)
+# =============================================================================
+BW:
+  name: "Baden-Wuerttemberg"
+  sources:
+    - domain: "km-bw.de"
+      name: "Kultusministerium Baden-Wuerttemberg"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: KM Baden-Wuerttemberg, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "bildungsplaene-bw.de"
+      name: "Bildungsplaene BW"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: Bildungsplan BW, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "schule-bw.de"
+      name: "Landesbildungsserver BW"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: Landesbildungsserver BW, {title}, {date}"
+      trust_boost: 0.85
+
+# =============================================================================
+# NORDRHEIN-WESTFALEN (NW)
+# =============================================================================
+NW:
+  name: "Nordrhein-Westfalen"
+  sources:
+    - domain: "schulministerium.nrw"
+      name: "Schulministerium NRW"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: MSB NRW, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "schulentwicklung.nrw.de"
+      name: "QUA-LiS NRW"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: QUA-LiS NRW, {title}, {date}"
+      trust_boost: 0.85
+
+    - domain: "learn-line.nrw.de"
+      name: "EDMOND NRW"
+      license: "CC-BY-SA"
+      legal_basis: "Creative Commons"
+      citation_template: "Quelle: EDMOND NRW, {title}, {date}"
+      trust_boost: 0.75
+
+# =============================================================================
+# HESSEN (HE)
+# =============================================================================
+HE:
+  name: "Hessen"
+  sources:
+    - domain: "kultusministerium.hessen.de"
+      name: "Kultusministerium Hessen"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: HKM Hessen, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "lsa.hessen.de"
+      name: "Landesschulamt Hessen"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: LSA Hessen, {title}, {date}"
+      trust_boost: 0.85
+
+    - domain: "bildung.hessen.de"
+      name: "Bildungsserver Hessen"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: Bildungsserver Hessen, {title}, {date}"
+      trust_boost: 0.85
+
+# =============================================================================
+# SACHSEN (SN)
+# =============================================================================
+SN:
+  name: "Sachsen"
+  sources:
+    - domain: "smk.sachsen.de"
+      name: "Kultusministerium Sachsen"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: SMK Sachsen, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "lehrplaene.sachsen.de"
+      name: "Lehrplaene Sachsen"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: Lehrplan Sachsen, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "sbi.smk.sachsen.de"
+      name: "SBI Sachsen"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: SBI Sachsen, {title}, {date}"
+      trust_boost: 0.85
+
+# =============================================================================
+# BERLIN (BE)
+# =============================================================================
+BE:
+  name: "Berlin"
+  sources:
+    - domain: "berlin.de/sen/bildung"
+      name: "Senatsverwaltung fuer Bildung Berlin"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: SenBJF Berlin, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "bildungsserver.berlin-brandenburg.de"
+      name: "Bildungsserver Berlin-Brandenburg"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: Bildungsserver Berlin-Brandenburg, {title}, {date}"
+      trust_boost: 0.85
+
+# =============================================================================
+# HAMBURG (HH)
+# =============================================================================
+HH:
+  name: "Hamburg"
+  sources:
+    - domain: "hamburg.de/bsb"
+      name: "Schulbehoerde Hamburg"
+      license: "§5 UrhG"
+      legal_basis: "Amtliche Werke (§5 UrhG)"
+      citation_template: "Quelle: BSB Hamburg, {title}, {date}"
+      trust_boost: 0.90
+
+    - domain: "li.hamburg.de"
+      name: "Landesinstitut Hamburg"
+      license: "DL-DE-BY-2.0"
+      legal_basis: "Datenlizenz Deutschland"
+      citation_template: "Quelle: LI Hamburg, {title}, {date}"
+      trust_boost: 0.85
+
+# =============================================================================
+# DEFAULT OPERATIONS MATRIX
+# =============================================================================
+# IMPORTANT: Training is ALWAYS forbidden!
+default_operations:
+  lookup:
+    allowed: true
+    requires_citation: true
+  rag:
+    allowed: true
+    requires_citation: true
+  training:
+    allowed: false  # VERBOTEN - Training with external data is NOT allowed
+  export:
+    allowed: true
+    requires_citation: true
+
+# =============================================================================
+# PII DETECTION RULES
+# =============================================================================
+pii_rules:
+  # Email Addresses
+  - name: "Email Addresses"
+    type: "regex"
+    pattern: "[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\\.[a-zA-Z]{2,}"
+    severity: "block"
+
+  # German Phone Numbers
+  - name: "German Phone Numbers"
+    type: "regex"
+    pattern: "(?:\\+49|0)[\\s.-]?\\d{2,4}[\\s.-]?\\d{3,}[\\s.-]?\\d{2,}"
+    severity: "block"
+
+  # German Mobile Numbers
+  - name: "German Mobile Numbers"
+    type: "regex"
+    pattern: "(?:\\+49|0)1[567]\\d[\\s.-]?\\d{3,}[\\s.-]?\\d{2,}"
+    severity: "block"
+
+  # IBAN (German)
+  - name: "German IBAN"
+    type: "regex"
+    pattern: "DE\\d{2}\\s?\\d{4}\\s?\\d{4}\\s?\\d{4}\\s?\\d{4}\\s?\\d{2}"
+    severity: "block"
+
+  # German Tax ID (Steuer-ID)
+  - name: "German Tax ID"
+    type: "regex"
+    pattern: "\\d{2}\\s?\\d{3}\\s?\\d{3}\\s?\\d{3}"
+    severity: "block"
+
+  # Credit Card Numbers
+  - name: "Credit Card Numbers"
+    type: "regex"
+    pattern: "(?:\\d{4}[\\s.-]?){3}\\d{4}"
+    severity: "block"
+
+  # German Address Pattern (Postal Code + City)
+  - name: "German Address Pattern"
+    type: "regex"
+    pattern: "\\d{5}\\s+[A-ZÄÖÜ][a-zäöüß]+"
+    severity: "warn"
+
+  # Date of Birth Patterns
+  - name: "Date of Birth"
+    type: "regex"
+    pattern: "(?:geboren|geb\\.|Geburtsdatum|DoB)[\\s:]*\\d{1,2}[\\./]\\d{1,2}[\\./]\\d{2,4}"
+    severity: "warn"
+
+  # Personal Names with Titles
+  - name: "Personal Names with Titles"
+    type: "regex"
+    pattern: "(?:Herr|Frau|Dr\\.|Prof\\.)\\s+[A-ZÄÖÜ][a-zäöüß]+\\s+[A-ZÄÖÜ][a-zäöüß]+"
+    severity: "warn"
+
+  # German Health Insurance Number
+  - name: "Health Insurance Number"
+    type: "regex"
+    pattern: "[A-Z]\\d{9}"
+    severity: "block"
--- a/edu-search-service/rules/doc_type_rules.yaml
+++ b/edu-search-service/rules/doc_type_rules.yaml
@@ -0,0 +1,178 @@
+# Doc-Type Klassifikationsregeln
+# Scoring-basiert: höchster Score gewinnt
+
+doc_types:
+  Lehrplan:
+    strong_terms:  # +4 pro Match
+      - "lehrplan"
+      - "kompetenzerwartungen"
+      - "fachanforderungen"
+      - "bildungsplan"
+      - "kerncurriculum"
+      - "rahmenlehrplan"
+      - "schulcurriculum"
+    medium_terms:  # +3 pro Match
+      - "kompetenzorientiert"
+      - "kompetenzbereiche"
+      - "inhaltsfelder"
+      - "anforderungsbereiche"
+      - "bildungsstandards"
+      - "stundentafel"
+    url_patterns:  # +2 pro Match
+      - "/lehrplan"
+      - "/curriculum"
+      - "/kerncurriculum"
+      - "/rahmenlehrplan"
+      - "/bildungsplan"
+
+  Kerncurriculum:
+    strong_terms:
+      - "kerncurriculum"
+      - "kc "
+    medium_terms:
+      - "prozessbezogene kompetenzen"
+      - "inhaltsbezogene kompetenzen"
+    url_patterns:
+      - "/kerncurriculum"
+      - "/kc/"
+
+  Rahmenlehrplan:
+    strong_terms:
+      - "rahmenlehrplan"
+      - "rlp"
+    url_patterns:
+      - "/rahmenlehrplan"
+      - "/rlp/"
+
+  Erlass_Verordnung:
+    strong_terms:  # +5 - rechtliche Texte haben Vorrang
+      - "erlass"
+      - "verordnung"
+      - "vo"
+      - "richtlinie"
+      - "amtsblatt"
+      - "gvbl"
+      - "gesetz"
+    medium_terms:
+      - "tritt in kraft"
+      - "gilt ab"
+      - "ausfertigungsdatum"
+      - "bekanntmachung"
+    url_patterns:
+      - "/amtsblatt"
+      - "/recht/"
+      - "/verordnungen"
+      - "/erlasse"
+      - "/bekanntmachung"
+
+  Arbeitsblatt:
+    strong_terms:
+      - "arbeitsblatt"
+      - "worksheet"
+      - "kopiervorlage"
+      - "loesungsblatt"
+      - "stationenlernen"
+      - "lerntheke"
+    medium_terms:
+      - "aufgabe 1"
+      - "aufgabe 2"
+      - "name:"
+      - "datum:"
+      - "klasse:"
+    url_patterns:
+      - "/arbeitsblatt"
+      - "/material/arbeitsblatt"
+      - "/download/arbeitsblatt"
+      - "/worksheet"
+
+  Unterrichtsentwurf:
+    strong_terms:
+      - "unterrichtsentwurf"
+      - "stundenentwurf"
+      - "verlaufsplan"
+      - "unterrichtsplanung"
+    medium_terms:
+      - "lernziele"
+      - "kompetenzziel"
+      - "einstieg"
+      - "sicherung"
+      - "transfer"
+      - "didaktische analyse"
+      - "methodische analyse"
+    url_patterns:
+      - "/unterrichtsentwurf"
+      - "/stundenentwurf"
+
+  Materialsammlung:
+    strong_terms:
+      - "materialsammlung"
+      - "materialpaket"
+      - "unterrichtsmaterial"
+    medium_terms:
+      - "materialien"
+      - "sammlung"
+      - "paket"
+    url_patterns:
+      - "/material"
+      - "/materialsammlung"
+
+  Pruefung_Abitur:
+    strong_terms:
+      - "abitur"
+      - "zentralabitur"
+      - "pruefungsaufgaben"
+      - "pruefung"
+      - "klausur"
+      - "aufgabenpool"
+    medium_terms:
+      - "hilfsmittel"
+      - "bearbeitungszeit"
+      - "bewertungshinweise"
+      - "loesungsvorschlag"
+    url_patterns:
+      - "/abitur"
+      - "/pruefung"
+      - "/klausur"
+
+  Studie_Bericht:
+    strong_terms:
+      - "studie"
+      - "bericht"
+      - "evaluation"
+      - "monitoring"
+      - "report"
+    medium_terms:
+      - "ergebnisse"
+      - "methodik"
+      - "stichprobe"
+      - "fragebogen"
+      - "datenanalyse"
+    url_patterns:
+      - "/studie"
+      - "/bericht"
+      - "/evaluation"
+
+  News_Blog:
+    strong_terms:
+      - "pressemitteilung"
+      - "aktuelles"
+    url_patterns:
+      - "/news"
+      - "/blog"
+      - "/presse"
+      - "/aktuelles"
+      - "/meldung"
+
+# Konfliktauflösung
+priority_order:
+  - Erlass_Verordnung  # Rechtliche Texte immer zuerst
+  - Pruefung_Abitur
+  - Lehrplan
+  - Kerncurriculum
+  - Rahmenlehrplan
+  - Arbeitsblatt
+  - Unterrichtsentwurf
+  - Materialsammlung
+  - Studie_Bericht
+  - News_Blog
+  - Sonstiges
--- a/edu-search-service/rules/level_rules.yaml
+++ b/edu-search-service/rules/level_rules.yaml
@@ -0,0 +1,121 @@
+# School Level Tagging Regeln
+
+levels:
+  Primar:
+    strong:
+      - "grundschule"
+      - "primarstufe"
+      - "klasse 1"
+      - "klasse 2"
+      - "klasse 3"
+      - "klasse 4"
+      - "1. klasse"
+      - "2. klasse"
+      - "3. klasse"
+      - "4. klasse"
+      - "sachunterricht"
+      - "schuleingangsphase"
+    weak:
+      - "anfangsunterricht"
+      - "schreibenlernen"
+      - "erstlesen"
+    negative: []
+
+  SekI:
+    strong:
+      - "sekundarstufe i"
+      - "sek i"
+      - "klasse 5"
+      - "klasse 6"
+      - "klasse 7"
+      - "klasse 8"
+      - "klasse 9"
+      - "klasse 10"
+      - "jahrgang 5"
+      - "jahrgang 6"
+      - "jahrgang 7"
+      - "jahrgang 8"
+      - "jahrgang 9"
+      - "jahrgang 10"
+      - "mittlere schule"
+      - "realschule"
+      - "hauptschule"
+      - "mittelschule"
+      - "erprobungsstufe"
+    weak:
+      - "5. klasse"
+      - "6. klasse"
+      - "7. klasse"
+      - "8. klasse"
+      - "9. klasse"
+      - "10. klasse"
+    negative: []
+
+  SekII:
+    strong:
+      - "sekundarstufe ii"
+      - "sek ii"
+      - "oberstufe"
+      - "gymnasiale oberstufe"
+      - "ef"
+      - "q1"
+      - "q2"
+      - "11. klasse"
+      - "12. klasse"
+      - "13. klasse"
+      - "jahrgang 11"
+      - "jahrgang 12"
+      - "jahrgang 13"
+      - "abitur"
+      - "abiturvorbereitung"
+      - "leistungskurs"
+      - "grundkurs"
+      - "qualifikationsphase"
+      - "einfuehrungsphase"
+    weak:
+      - "oberstufenschueler"
+      - "kursstufe"
+    negative: []
+
+  Beruf:
+    strong:
+      - "berufsschule"
+      - "ausbildung"
+      - "ihk"
+      - "lernfeld"
+      - "berufliches gymnasium"
+      - "berufskolleg"
+      - "berufsfachschule"
+      - "fachschule"
+      - "duales system"
+      - "auszubildende"
+    weak:
+      - "betrieb"
+      - "praxis"
+    negative: []
+
+  Hochschule:
+    strong:
+      - "modulhandbuch"
+      - "ects"
+      - "seminar universitaet"
+      - "vorlesung"
+      - "studiengang"
+      - "bachelor"
+      - "master"
+      - "dissertation"
+      - "hochschuldidaktik"
+    weak:
+      - "studierende"
+      - "hochschule"
+    negative: []
+
+# Konfliktauflösung
+# Wenn mehrere Levels matchen, gewinnt der spezifischere
+priority_order:
+  - Primar
+  - SekI
+  - SekII
+  - Beruf
+  - Hochschule
+  - NA
--- a/edu-search-service/rules/subject_rules.yaml
+++ b/edu-search-service/rules/subject_rules.yaml
@@ -0,0 +1,285 @@
+# Subject (Fach) Tagging Regeln
+# Format: strong (+3), weak (+1), negative (-2)
+
+subjects:
+  Mathe:
+    strong:
+      - "mathematik"
+      - "mathe"
+      - "algebra"
+      - "geometrie"
+      - "stochastik"
+      - "analysis"
+      - "prozentrechnung"
+      - "gleichungen"
+      - "funktionen"
+      - "trigonometrie"
+      - "wahrscheinlichkeit"
+    weak:
+      - "zahlen"
+      - "terme"
+      - "diagramme"
+      - "brueche"
+      - "dreisatz"
+      - "rechnen"
+    negative: []
+
+  Deutsch:
+    strong:
+      - "deutschunterricht"
+      - "grammatik"
+      - "rechtschreibung"
+      - "aufsatz"
+      - "textanalyse"
+      - "literatur"
+      - "argumentation"
+      - "erzaehlung"
+      - "lyrik"
+      - "drama"
+      - "epik"
+    weak:
+      - "lesen"
+      - "schreiben"
+      - "woerter"
+      - "satzglieder"
+    negative:
+      - "deutschland"
+      - "deutsche geschichte"
+
+  Englisch:
+    strong:
+      - "englischunterricht"
+      - "english"
+      - "grammar"
+      - "vocabulary"
+      - "reading comprehension"
+      - "listening"
+      - "speaking"
+    weak:
+      - "text"
+      - "dialogue"
+    negative: []
+
+  Franzoesisch:
+    strong:
+      - "franzoesisch"
+      - "francais"
+      - "french"
+      - "grammaire"
+    weak:
+      - "texte"
+    negative: []
+
+  Latein:
+    strong:
+      - "latein"
+      - "lateinunterricht"
+      - "grammatik latein"
+    weak:
+      - "uebersetzung"
+      - "vokabel"
+    negative: []
+
+  Biologie:
+    strong:
+      - "biologie"
+      - "bio"
+      - "oekologie"
+      - "evolution"
+      - "genetik"
+      - "zellbiologie"
+      - "stoffwechsel"
+      - "neurobiologie"
+    weak:
+      - "zelle"
+      - "organismus"
+      - "lebewesen"
+    negative: []
+
+  Chemie:
+    strong:
+      - "chemie"
+      - "chemieunterricht"
+      - "organische chemie"
+      - "anorganische chemie"
+      - "reaktionsgleichung"
+      - "periodensystem"
+    weak:
+      - "element"
+      - "verbindung"
+      - "reaktion"
+    negative: []
+
+  Physik:
+    strong:
+      - "physik"
+      - "physikunterricht"
+      - "mechanik"
+      - "elektrizitaet"
+      - "optik"
+      - "thermodynamik"
+      - "quantenphysik"
+    weak:
+      - "energie"
+      - "kraft"
+      - "bewegung"
+    negative: []
+
+  Informatik:
+    strong:
+      - "informatik"
+      - "programmierung"
+      - "algorithmus"
+      - "datenstruktur"
+      - "python"
+      - "java"
+      - "sql"
+      - "netzwerke"
+    weak:
+      - "code"
+      - "daten"
+      - "computer"
+    negative: []
+
+  Geschichte:
+    strong:
+      - "geschichtsunterricht"
+      - "historisch"
+      - "weimarer republik"
+      - "nationalsozialismus"
+      - "mittelalter"
+      - "aufklaerung"
+      - "industrialisierung"
+      - "antike"
+      - "renaissance"
+    weak:
+      - "quelle"
+      - "chronologie"
+      - "epoche"
+    negative: []
+
+  Politik_Sozialkunde:
+    strong:
+      - "politik"
+      - "politikunterricht"
+      - "sozialkunde"
+      - "gemeinschaftskunde"
+      - "demokratie"
+      - "grundgesetz"
+      - "bundestag"
+      - "wahlen"
+      - "parteien"
+    weak:
+      - "rechte"
+      - "pflichten"
+      - "institutionen"
+    negative:
+      - "europaeische union" # zu allgemein
+
+  Geographie:
+    strong:
+      - "geographie"
+      - "geografie"
+      - "erdkunde"
+      - "topographie"
+      - "klimazonen"
+      - "plattentektonik"
+    weak:
+      - "karte"
+      - "landschaft"
+      - "kontinent"
+    negative: []
+
+  Religion_Ethik:
+    strong:
+      - "religionsunterricht"
+      - "ethik"
+      - "philosophie"
+      - "weltreligionen"
+      - "bibel"
+      - "christentum"
+      - "islam"
+      - "judentum"
+    weak:
+      - "werte"
+      - "moral"
+    negative: []
+
+  Kunst:
+    strong:
+      - "kunstunterricht"
+      - "bildende kunst"
+      - "malerei"
+      - "zeichnen"
+      - "gestaltung"
+      - "kunstgeschichte"
+    weak:
+      - "bild"
+      - "farbe"
+    negative:
+      - "kunststoff"
+
+  Musik:
+    strong:
+      - "musikunterricht"
+      - "musiktheorie"
+      - "notenlehre"
+      - "rhythmus"
+      - "harmonie"
+      - "instrument"
+    weak:
+      - "lied"
+      - "melodie"
+    negative: []
+
+  Sport:
+    strong:
+      - "sportunterricht"
+      - "bewegung sport"
+      - "leichtathletik"
+      - "ballsport"
+      - "turnen"
+      - "schwimmen unterricht"
+    weak:
+      - "spiel"
+      - "fitness"
+    negative:
+      - "sportlich"
+      - "esport"
+
+  Wirtschaft:
+    strong:
+      - "wirtschaftsunterricht"
+      - "oekonomie"
+      - "volkswirtschaft"
+      - "betriebswirtschaft"
+      - "marktwirtschaft"
+    weak:
+      - "unternehmen"
+      - "markt"
+    negative: []
+
+  Sachunterricht:
+    strong:
+      - "sachunterricht"
+      - "heimat- und sachunterricht"
+      - "hsu"
+    weak:
+      - "grundschule thema"
+    negative: []
+
+  DaZ_DaF:
+    strong:
+      - "deutsch als zweitsprache"
+      - "deutsch als fremdsprache"
+      - "daz"
+      - "daf"
+      - "alphabetisierung"
+    weak:
+      - "sprachfoerderung"
+      - "integration"
+    negative: []
+
+# Threshold für Subject-Assignment
+threshold: 4  # Mindest-Score um Subject zuzuweisen
+max_subjects: 3  # Maximal 3 Subjects pro Dokument
--- a/edu-search-service/rules/trust_rules.yaml
+++ b/edu-search-service/rules/trust_rules.yaml
@@ -0,0 +1,117 @@
+# Trust Score Regeln für Education Search
+# Score-Berechnung: Summe aller matchenden Regeln, dann clamp(0, 1)
+
+domain_boosts:
+  # Bundesebene (höchste Vertrauensstufe)
+  - match: "*.kmk.org"
+    add: 0.50
+    reason: "KMK - Kultusministerkonferenz"
+  - match: "*.bildungsserver.de"
+    add: 0.50
+    reason: "Deutscher Bildungsserver"
+  - match: "*.bpb.de"
+    add: 0.45
+    reason: "Bundeszentrale für politische Bildung"
+  - match: "*.bmbf.de"
+    add: 0.50
+    reason: "BMBF"
+  - match: "*.iqb.hu-berlin.de"
+    add: 0.50
+    reason: "IQB Bildungsstandards"
+
+  # Landesministerien
+  - match: "*.bayern.de"
+    add: 0.45
+    reason: "Bayern offiziell"
+  - match: "*.nrw.de"
+    add: 0.45
+    reason: "NRW offiziell"
+  - match: "*.berlin.de"
+    add: 0.45
+    reason: "Berlin offiziell"
+  - match: "*.sachsen.de"
+    add: 0.45
+    reason: "Sachsen offiziell"
+  - match: "*.niedersachsen.de"
+    add: 0.45
+    reason: "Niedersachsen offiziell"
+  - match: "*.hessen.de"
+    add: 0.45
+    reason: "Hessen offiziell"
+  - match: "*.brandenburg.de"
+    add: 0.45
+    reason: "Brandenburg offiziell"
+  - match: "*.thueringen.de"
+    add: 0.45
+    reason: "Thüringen offiziell"
+
+  # Bildungsserver der Länder
+  - match: "*.nibis.de"
+    add: 0.40
+    reason: "Niedersachsen Bildungsserver"
+  - match: "*.learnline.nrw.de"
+    add: 0.40
+    reason: "NRW Bildungsserver"
+  - match: "*.schule-bw.de"
+    add: 0.40
+    reason: "BW Bildungsserver"
+
+  # Universitäten
+  - match: "*.uni-*.de"
+    add: 0.30
+    reason: "Deutsche Universität"
+  - match: "*.tu-*.de"
+    add: 0.30
+    reason: "Technische Universität"
+  - match: "*.fh-*.de"
+    add: 0.25
+    reason: "Fachhochschule"
+
+  # Etablierte Portale
+  - match: "*.zum.de"
+    add: 0.25
+    reason: "ZUM - Zentrale für Unterrichtsmedien"
+  - match: "*.lehrer-online.de"
+    add: 0.20
+    reason: "Lehrer-Online Portal"
+  - match: "*.4teachers.de"
+    add: 0.20
+    reason: "4teachers Portal"
+
+tld_boosts:
+  - tld: ".gov"
+    add: 0.40
+    reason: "Government TLD"
+  - tld: ".edu"
+    add: 0.35
+    reason: "Education TLD"
+
+penalties:
+  # URL-Muster die Werbung/Tracking andeuten
+  - if_url_contains: ["utm_", "affiliate", "partner=", "ref="]
+    add: -0.10
+    reason: "Tracking/Affiliate Parameter"
+
+  # Kommerzielle Signale
+  - if_url_contains: ["shop", "kaufen", "bestellen", "warenkorb"]
+    add: -0.20
+    reason: "E-Commerce Signale"
+
+  # SEO-Spam Indikatoren
+  - if_url_contains: ["gratis-", "kostenlos-download", "sofort-"]
+    add: -0.15
+    reason: "SEO-Spam Muster"
+
+# Content-basierte Strafen (werden vom Extractor gesetzt)
+content_penalties:
+  - if_ad_density_gt: 0.20
+    add: -0.30
+    reason: "Hoher Werbeanteil"
+
+  - if_link_density_gt: 0.40
+    add: -0.20
+    reason: "Hohe Link-Dichte (Linkfarm)"
+
+  - if_content_length_lt: 200
+    add: -0.25
+    reason: "Sehr wenig Content"
--- a/edu-search-service/scripts/add_german_universities.py
+++ b/edu-search-service/scripts/add_german_universities.py
@@ -0,0 +1,282 @@
+#!/usr/bin/env python3
+"""
+Add all major German universities to the edu-search-service database.
+Based on HRK (Hochschulrektorenkonferenz) list.
+"""
+
+import requests
+import json
+import time
+import sys
+
+API_BASE = "https://macmini:8089/api/v1"
+
+# German Universities - categorized
+GERMAN_UNIVERSITIES = [
+    # === UNIVERSITIES (Universitäten) ===
+    # Already in DB (skip or update)
+    # {"name": "TUM", "url": "https://www.tum.de", "type": "university"},
+    # {"name": "LMU München", "url": "https://www.lmu.de", "type": "university"},
+    # {"name": "UOL", "url": "https://uol.de", "type": "university"},
+    # {"name": "KIT Karlsruhe", "url": "https://www.kit.edu", "type": "university"},
+
+    # TU9 Universities
+    {"name": "TU Dresden", "url": "https://tu-dresden.de", "type": "university"},
+    {"name": "TU Braunschweig", "url": "https://www.tu-braunschweig.de", "type": "university"},
+    {"name": "TU Darmstadt", "url": "https://www.tu-darmstadt.de", "type": "university"},
+    {"name": "Leibniz Universität Hannover", "url": "https://www.uni-hannover.de", "type": "university"},
+    {"name": "Universität Stuttgart", "url": "https://www.uni-stuttgart.de", "type": "university"},
+
+    # Excellence Universities
+    {"name": "Universität Bonn", "url": "https://www.uni-bonn.de", "type": "university"},
+    {"name": "Universität Konstanz", "url": "https://www.uni-konstanz.de", "type": "university"},
+    {"name": "Universität Tübingen", "url": "https://uni-tuebingen.de", "type": "university"},
+    {"name": "Universität Freiburg", "url": "https://www.uni-freiburg.de", "type": "university"},
+
+    # Large State Universities
+    {"name": "Universität Münster", "url": "https://www.uni-muenster.de", "type": "university"},
+    {"name": "Universität Frankfurt", "url": "https://www.uni-frankfurt.de", "type": "university"},
+    {"name": "Universität Mainz", "url": "https://www.uni-mainz.de", "type": "university"},
+    {"name": "Universität Würzburg", "url": "https://www.uni-wuerzburg.de", "type": "university"},
+    {"name": "Universität Erlangen-Nürnberg", "url": "https://www.fau.de", "type": "university"},
+    {"name": "Universität Leipzig", "url": "https://www.uni-leipzig.de", "type": "university"},
+    {"name": "Universität Jena", "url": "https://www.uni-jena.de", "type": "university"},
+    {"name": "Universität Halle", "url": "https://www.uni-halle.de", "type": "university"},
+    {"name": "Universität Rostock", "url": "https://www.uni-rostock.de", "type": "university"},
+    {"name": "Universität Greifswald", "url": "https://www.uni-greifswald.de", "type": "university"},
+    {"name": "Universität Kiel", "url": "https://www.uni-kiel.de", "type": "university"},
+    {"name": "Universität Bremen", "url": "https://www.uni-bremen.de", "type": "university"},
+    {"name": "Universität Bielefeld", "url": "https://www.uni-bielefeld.de", "type": "university"},
+    {"name": "Universität Duisburg-Essen", "url": "https://www.uni-due.de", "type": "university"},
+    {"name": "Universität Dortmund", "url": "https://www.tu-dortmund.de", "type": "university"},
+    {"name": "Universität Bochum", "url": "https://www.ruhr-uni-bochum.de", "type": "university"},
+    {"name": "Universität Düsseldorf", "url": "https://www.hhu.de", "type": "university"},
+    {"name": "Universität Wuppertal", "url": "https://www.uni-wuppertal.de", "type": "university"},
+    {"name": "Universität Siegen", "url": "https://www.uni-siegen.de", "type": "university"},
+    {"name": "Universität Paderborn", "url": "https://www.uni-paderborn.de", "type": "university"},
+    {"name": "Universität Kassel", "url": "https://www.uni-kassel.de", "type": "university"},
+    {"name": "Universität Marburg", "url": "https://www.uni-marburg.de", "type": "university"},
+    {"name": "Universität Gießen", "url": "https://www.uni-giessen.de", "type": "university"},
+    {"name": "Universität Saarbrücken", "url": "https://www.uni-saarland.de", "type": "university"},
+    {"name": "Universität Trier", "url": "https://www.uni-trier.de", "type": "university"},
+    {"name": "Universität Koblenz", "url": "https://www.uni-koblenz.de", "type": "university"},
+    {"name": "Universität Landau", "url": "https://rptu.de", "type": "university"},
+    {"name": "Universität Mannheim", "url": "https://www.uni-mannheim.de", "type": "university"},
+    {"name": "Universität Ulm", "url": "https://www.uni-ulm.de", "type": "university"},
+    {"name": "Universität Hohenheim", "url": "https://www.uni-hohenheim.de", "type": "university"},
+    {"name": "Universität Regensburg", "url": "https://www.uni-regensburg.de", "type": "university"},
+    {"name": "Universität Passau", "url": "https://www.uni-passau.de", "type": "university"},
+    {"name": "Universität Bayreuth", "url": "https://www.uni-bayreuth.de", "type": "university"},
+    {"name": "Universität Bamberg", "url": "https://www.uni-bamberg.de", "type": "university"},
+    {"name": "Universität Augsburg", "url": "https://www.uni-augsburg.de", "type": "university"},
+    {"name": "Universität Potsdam", "url": "https://www.uni-potsdam.de", "type": "university"},
+    {"name": "Universität Magdeburg", "url": "https://www.ovgu.de", "type": "university"},
+    {"name": "TU Chemnitz", "url": "https://www.tu-chemnitz.de", "type": "university"},
+    {"name": "TU Ilmenau", "url": "https://www.tu-ilmenau.de", "type": "university"},
+    {"name": "TU Freiberg", "url": "https://tu-freiberg.de", "type": "university"},
+    {"name": "TU Clausthal", "url": "https://www.tu-clausthal.de", "type": "university"},
+    {"name": "TU Kaiserslautern", "url": "https://rptu.de", "type": "university"},
+    {"name": "BTU Cottbus-Senftenberg", "url": "https://www.b-tu.de", "type": "university"},
+    {"name": "Universität der Bundeswehr München", "url": "https://www.unibw.de", "type": "university"},
+    {"name": "Universität der Bundeswehr Hamburg", "url": "https://www.hsu-hh.de", "type": "university"},
+
+    # === FACHHOCHSCHULEN / HAW ===
+    {"name": "HAW Hamburg", "url": "https://www.haw-hamburg.de", "type": "haw"},
+    {"name": "HTW Berlin", "url": "https://www.htw-berlin.de", "type": "haw"},
+    {"name": "Beuth Hochschule Berlin", "url": "https://www.bht-berlin.de", "type": "haw"},
+    {"name": "FH Aachen", "url": "https://www.fh-aachen.de", "type": "haw"},
+    {"name": "TH Köln", "url": "https://www.th-koeln.de", "type": "haw"},
+    {"name": "Hochschule Düsseldorf", "url": "https://www.hs-duesseldorf.de", "type": "haw"},
+    {"name": "FH Dortmund", "url": "https://www.fh-dortmund.de", "type": "haw"},
+    {"name": "Hochschule Bochum", "url": "https://www.hochschule-bochum.de", "type": "haw"},
+    {"name": "Westfälische Hochschule", "url": "https://www.w-hs.de", "type": "haw"},
+    {"name": "FH Bielefeld", "url": "https://www.fh-bielefeld.de", "type": "haw"},
+    {"name": "FH Münster", "url": "https://www.fh-muenster.de", "type": "haw"},
+    {"name": "Hochschule Osnabrück", "url": "https://www.hs-osnabrueck.de", "type": "haw"},
+    {"name": "Hochschule Bremen", "url": "https://www.hs-bremen.de", "type": "haw"},
+    {"name": "Hochschule Hannover", "url": "https://www.hs-hannover.de", "type": "haw"},
+    {"name": "Ostfalia Hochschule", "url": "https://www.ostfalia.de", "type": "haw"},
+    {"name": "Hochschule Emden/Leer", "url": "https://www.hs-emden-leer.de", "type": "haw"},
+    {"name": "HAWK Hildesheim", "url": "https://www.hawk.de", "type": "haw"},
+    {"name": "Hochschule Fulda", "url": "https://www.hs-fulda.de", "type": "haw"},
+    {"name": "Frankfurt UAS", "url": "https://www.frankfurt-university.de", "type": "haw"},
+    {"name": "Hochschule Darmstadt", "url": "https://www.h-da.de", "type": "haw"},
+    {"name": "Hochschule RheinMain", "url": "https://www.hs-rm.de", "type": "haw"},
+    {"name": "Hochschule Mainz", "url": "https://www.hs-mainz.de", "type": "haw"},
+    {"name": "Hochschule Trier", "url": "https://www.hochschule-trier.de", "type": "haw"},
+    {"name": "Hochschule Koblenz", "url": "https://www.hs-koblenz.de", "type": "haw"},
+    {"name": "Hochschule Karlsruhe", "url": "https://www.h-ka.de", "type": "haw"},
+    {"name": "Hochschule Mannheim", "url": "https://www.hs-mannheim.de", "type": "haw"},
+    {"name": "Hochschule Heilbronn", "url": "https://www.hs-heilbronn.de", "type": "haw"},
+    {"name": "Hochschule Esslingen", "url": "https://www.hs-esslingen.de", "type": "haw"},
+    {"name": "Hochschule Reutlingen", "url": "https://www.reutlingen-university.de", "type": "haw"},
+    {"name": "Hochschule Konstanz", "url": "https://www.htwg-konstanz.de", "type": "haw"},
+    {"name": "Hochschule Offenburg", "url": "https://www.hs-offenburg.de", "type": "haw"},
+    {"name": "Hochschule Pforzheim", "url": "https://www.hs-pforzheim.de", "type": "haw"},
+    {"name": "Hochschule Albstadt-Sigmaringen", "url": "https://www.hs-albsig.de", "type": "haw"},
+    {"name": "Hochschule München", "url": "https://www.hm.edu", "type": "haw"},
+    {"name": "TH Nürnberg", "url": "https://www.th-nuernberg.de", "type": "haw"},
+    {"name": "TH Ingolstadt", "url": "https://www.thi.de", "type": "haw"},
+    {"name": "Hochschule Augsburg", "url": "https://www.hs-augsburg.de", "type": "haw"},
+    {"name": "Hochschule Rosenheim", "url": "https://www.th-rosenheim.de", "type": "haw"},
+    {"name": "Hochschule Regensburg", "url": "https://www.oth-regensburg.de", "type": "haw"},
+    {"name": "Hochschule Landshut", "url": "https://www.haw-landshut.de", "type": "haw"},
+    {"name": "Hochschule Coburg", "url": "https://www.hs-coburg.de", "type": "haw"},
+    {"name": "Hochschule Hof", "url": "https://www.hof-university.de", "type": "haw"},
+    {"name": "Hochschule Würzburg-Schweinfurt", "url": "https://www.thws.de", "type": "haw"},
+    {"name": "Hochschule Aschaffenburg", "url": "https://www.th-ab.de", "type": "haw"},
+    {"name": "Hochschule Ansbach", "url": "https://www.hs-ansbach.de", "type": "haw"},
+    {"name": "OTH Amberg-Weiden", "url": "https://www.oth-aw.de", "type": "haw"},
+    {"name": "Hochschule Deggendorf", "url": "https://www.th-deg.de", "type": "haw"},
+    {"name": "Hochschule Kempten", "url": "https://www.hs-kempten.de", "type": "haw"},
+    {"name": "Hochschule Neu-Ulm", "url": "https://www.hnu.de", "type": "haw"},
+    {"name": "HTW Dresden", "url": "https://www.htw-dresden.de", "type": "haw"},
+    {"name": "HTWK Leipzig", "url": "https://www.htwk-leipzig.de", "type": "haw"},
+    {"name": "Hochschule Mittweida", "url": "https://www.hs-mittweida.de", "type": "haw"},
+    {"name": "Hochschule Zittau/Görlitz", "url": "https://www.hszg.de", "type": "haw"},
+    {"name": "Westsächsische Hochschule Zwickau", "url": "https://www.fh-zwickau.de", "type": "haw"},
+    {"name": "Hochschule Merseburg", "url": "https://www.hs-merseburg.de", "type": "haw"},
+    {"name": "Hochschule Anhalt", "url": "https://www.hs-anhalt.de", "type": "haw"},
+    {"name": "Hochschule Magdeburg-Stendal", "url": "https://www.h2.de", "type": "haw"},
+    {"name": "Hochschule Harz", "url": "https://www.hs-harz.de", "type": "haw"},
+    {"name": "Ernst-Abbe-Hochschule Jena", "url": "https://www.eah-jena.de", "type": "haw"},
+    {"name": "FH Erfurt", "url": "https://www.fh-erfurt.de", "type": "haw"},
+    {"name": "Hochschule Nordhausen", "url": "https://www.hs-nordhausen.de", "type": "haw"},
+    {"name": "Hochschule Schmalkalden", "url": "https://www.hs-schmalkalden.de", "type": "haw"},
+    {"name": "TH Brandenburg", "url": "https://www.th-brandenburg.de", "type": "haw"},
+    {"name": "FH Potsdam", "url": "https://www.fh-potsdam.de", "type": "haw"},
+    {"name": "TH Wildau", "url": "https://www.th-wildau.de", "type": "haw"},
+    {"name": "Hochschule Neubrandenburg", "url": "https://www.hs-nb.de", "type": "haw"},
+    {"name": "Hochschule Stralsund", "url": "https://www.hochschule-stralsund.de", "type": "haw"},
+    {"name": "Hochschule Wismar", "url": "https://www.hs-wismar.de", "type": "haw"},
+    {"name": "FH Kiel", "url": "https://www.fh-kiel.de", "type": "haw"},
+    {"name": "FH Westküste", "url": "https://www.fh-westkueste.de", "type": "haw"},
+    {"name": "TH Lübeck", "url": "https://www.th-luebeck.de", "type": "haw"},
+    {"name": "FH Flensburg", "url": "https://hs-flensburg.de", "type": "haw"},
+    {"name": "Hochschule Bremerhaven", "url": "https://www.hs-bremerhaven.de", "type": "haw"},
+
+    # === PRIVATE HOCHSCHULEN ===
+    {"name": "WHU Vallendar", "url": "https://www.whu.edu", "type": "private"},
+    {"name": "HHL Leipzig", "url": "https://www.hhl.de", "type": "private"},
+    {"name": "EBS Universität", "url": "https://www.ebs.edu", "type": "private"},
+    {"name": "Frankfurt School", "url": "https://www.frankfurt-school.de", "type": "private"},
+    {"name": "ESMT Berlin", "url": "https://esmt.berlin", "type": "private"},
+    {"name": "Jacobs University Bremen", "url": "https://www.jacobs-university.de", "type": "private"},
+    {"name": "Zeppelin Universität", "url": "https://www.zu.de", "type": "private"},
+    {"name": "Bucerius Law School", "url": "https://www.law-school.de", "type": "private"},
+    {"name": "Universität Witten/Herdecke", "url": "https://www.uni-wh.de", "type": "private"},
+    {"name": "IUBH", "url": "https://www.iu.de", "type": "private"},
+    {"name": "SRH Hochschule Heidelberg", "url": "https://www.srh-hochschule-heidelberg.de", "type": "private"},
+    {"name": "FOM Hochschule", "url": "https://www.fom.de", "type": "private"},
+
+    # === FRAUNHOFER INSTITUTE ===
+    {"name": "Fraunhofer IIS", "url": "https://www.iis.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IAIS", "url": "https://www.iais.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IML", "url": "https://www.iml.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer ISI", "url": "https://www.isi.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IPA", "url": "https://www.ipa.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IAO", "url": "https://www.iao.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IWS", "url": "https://www.iws.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IPT", "url": "https://www.ipt.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer FOKUS", "url": "https://www.fokus.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer HHI", "url": "https://www.hhi.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IESE", "url": "https://www.iese.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IOSB", "url": "https://www.iosb.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IDMT", "url": "https://www.idmt.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IKTS", "url": "https://www.ikts.fraunhofer.de", "type": "research"},
+    {"name": "Fraunhofer IGD", "url": "https://www.igd.fraunhofer.de", "type": "research"},
+
+    # === MAX-PLANCK-INSTITUTE ===
+    {"name": "MPI für Informatik", "url": "https://www.mpi-inf.mpg.de", "type": "research"},
+    {"name": "MPI für Software Systeme", "url": "https://www.mpi-sws.org", "type": "research"},
+    {"name": "MPI für intelligente Systeme", "url": "https://is.mpg.de", "type": "research"},
+    {"name": "MPI für Mathematik", "url": "https://www.mpim-bonn.mpg.de", "type": "research"},
+    {"name": "MPI für Physik", "url": "https://www.mpp.mpg.de", "type": "research"},
+    {"name": "MPI für Quantenoptik", "url": "https://www.mpq.mpg.de", "type": "research"},
+    {"name": "MPI für Biophysik", "url": "https://www.biophys.mpg.de", "type": "research"},
+    {"name": "MPI für Biochemie", "url": "https://www.biochem.mpg.de", "type": "research"},
+    {"name": "MPI für Neurobiologie", "url": "https://www.neuro.mpg.de", "type": "research"},
+    {"name": "MPI für Hirnforschung", "url": "https://brain.mpg.de", "type": "research"},
+
+    # === HELMHOLTZ-ZENTREN ===
+    {"name": "DESY Hamburg", "url": "https://www.desy.de", "type": "research"},
+    {"name": "FZ Jülich", "url": "https://www.fz-juelich.de", "type": "research"},
+    {"name": "GSI Darmstadt", "url": "https://www.gsi.de", "type": "research"},
+    {"name": "DKFZ Heidelberg", "url": "https://www.dkfz.de", "type": "research"},
+    {"name": "DLR", "url": "https://www.dlr.de", "type": "research"},
+    {"name": "AWI Bremerhaven", "url": "https://www.awi.de", "type": "research"},
+    {"name": "GFZ Potsdam", "url": "https://www.gfz-potsdam.de", "type": "research"},
+    {"name": "UFZ Leipzig", "url": "https://www.ufz.de", "type": "research"},
+    {"name": "GEOMAR Kiel", "url": "https://www.geomar.de", "type": "research"},
+]
+
+def get_existing_universities():
+    """Get list of existing universities from the API."""
+    try:
+        response = requests.get(f"{API_BASE}/universities", verify=False, timeout=10)
+        if response.status_code == 200:
+            data = response.json()
+            return {u['url'].rstrip('/').lower(): u for u in data.get('universities', [])}
+    except Exception as e:
+        print(f"Error fetching existing universities: {e}")
+    return {}
+
+def add_university(uni):
+    """Add a university to the database."""
+    payload = {
+        "name": uni["name"],
+        "url": uni["url"],
+        "type": uni.get("type", "university"),
+        "country": "DE"
+    }
+
+    try:
+        response = requests.post(
+            f"{API_BASE}/universities",
+            json=payload,
+            verify=False,
+            timeout=10
+        )
+        return response.status_code == 201 or response.status_code == 200
+    except Exception as e:
+        print(f"Error adding {uni['name']}: {e}")
+        return False
+
+def main():
+    print("Fetching existing universities...")
+    existing = get_existing_universities()
+    print(f"Found {len(existing)} existing universities")
+
+    added = 0
+    skipped = 0
+    failed = 0
+
+    for uni in GERMAN_UNIVERSITIES:
+        url_key = uni["url"].rstrip('/').lower()
+
+        if url_key in existing:
+            print(f"SKIP: {uni['name']} (already exists)")
+            skipped += 1
+            continue
+
+        print(f"ADD:  {uni['name']} ({uni['url']})")
+        if add_university(uni):
+            added += 1
+        else:
+            failed += 1
+
+        # Rate limiting
+        time.sleep(0.2)
+
+    print(f"\n=== SUMMARY ===")
+    print(f"Added:   {added}")
+    print(f"Skipped: {skipped}")
+    print(f"Failed:  {failed}")
+    print(f"Total:   {len(GERMAN_UNIVERSITIES)}")
+
+if __name__ == "__main__":
+    # Disable SSL warnings for self-signed cert
+    import urllib3
+    urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
+
+    main()
--- a/edu-search-service/scripts/fix_university_types.py
+++ b/edu-search-service/scripts/fix_university_types.py
@@ -0,0 +1,125 @@
+#!/usr/bin/env python3
+"""
+Fix university types in the database.
+This script updates uni_type based on university names.
+"""
+
+import requests
+import json
+import sys
+
+API_BASE = "https://macmini:8089/api/v1"
+
+# Classification rules based on name patterns
+UNI_TYPE_RULES = {
+    "UNI": [
+        "Universität", "University", "TU ", "TUM", "LMU", "RWTH",
+        "Humboldt", "FU Berlin", "HU Berlin", "TH ", "KIT"
+    ],
+    "FH": [
+        "Hochschule", "Fachhochschule", "FH ", "HAW ", "HS ",
+        "University of Applied", "Beuth", "HTW"
+    ],
+    "RESEARCH": [
+        "Fraunhofer", "Max-Planck", "Helmholtz", "DLR", "DESY",
+        "DKFZ", "FZ Jülich", "AWI", "GFZ", "GSI", "Leibniz"
+    ],
+    "PRIVATE": [
+        "EBS", "ESMT", "Bucerius", "WHU", "HHL", "FOM", "IUBH",
+        "SRH", "International School", "Business School"
+    ],
+    "KUNST": [
+        "Kunsthochschule", "Musikhochschule", "Filmhochschule",
+        "Kunstakademie", "HfK", "HfM", "HfG", "UdK", "Bauhaus"
+    ],
+    "PH": [
+        "Pädagogische Hochschule", "PH "
+    ]
+}
+
+def classify_university(name):
+    """Classify university by name patterns."""
+    name_lower = name.lower()
+
+    # Check each category
+    for uni_type, patterns in UNI_TYPE_RULES.items():
+        for pattern in patterns:
+            if pattern.lower() in name_lower:
+                return uni_type
+
+    # Default to UNI if "universität" in name, else FH
+    if "universität" in name_lower or "university" in name_lower:
+        return "UNI"
+
+    return "FH"  # Default
+
+def get_all_universities():
+    """Get list of all universities from the API."""
+    try:
+        response = requests.get(f"{API_BASE}/universities", verify=False, timeout=30)
+        if response.status_code == 200:
+            data = response.json()
+            return data.get('universities', [])
+    except Exception as e:
+        print(f"Error fetching universities: {e}")
+    return []
+
+def update_university_type(uni_id, uni_type, uni_state=None):
+    """Update university type via direct database or API."""
+    # The API doesn't have an update endpoint, so we'll print SQL statements
+    return uni_type
+
+def main():
+    print("=== University Type Fixer ===\n")
+
+    # Disable SSL warnings
+    import urllib3
+    urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
+
+    universities = get_all_universities()
+    if not universities:
+        print("ERROR: No universities found!")
+        return
+
+    print(f"Found {len(universities)} universities\n")
+
+    # Classify and generate SQL
+    sql_statements = []
+    type_counts = {}
+
+    for uni in universities:
+        uni_id = uni['id']
+        uni_name = uni['name']
+        current_type = uni.get('uni_type', 'unknown')
+
+        # Classify
+        new_type = classify_university(uni_name)
+
+        # Count
+        type_counts[new_type] = type_counts.get(new_type, 0) + 1
+
+        # Generate SQL
+        sql = f"UPDATE universities SET uni_type = '{new_type}' WHERE id = '{uni_id}';"
+        sql_statements.append(sql)
+
+        if current_type != new_type:
+            print(f"  {uni_name[:50]:<50} -> {new_type}")
+
+    print(f"\n=== Summary ===")
+    for t, c in sorted(type_counts.items()):
+        print(f"  {t}: {c}")
+
+    # Write SQL file
+    sql_file = "/tmp/fix_uni_types.sql"
+    with open(sql_file, 'w') as f:
+        f.write("-- Fix university types\n")
+        f.write("BEGIN;\n\n")
+        for sql in sql_statements:
+            f.write(sql + "\n")
+        f.write("\nCOMMIT;\n")
+
+    print(f"\nSQL written to: {sql_file}")
+    print(f"Run: cat {sql_file} | docker exec -i breakpilot-pwa-postgres psql -U <user> -d edu_search")
+
+if __name__ == "__main__":
+    main()
--- a/edu-search-service/scripts/seed_universities.py
+++ b/edu-search-service/scripts/seed_universities.py
@@ -0,0 +1,147 @@
+#!/usr/bin/env python3
+"""
+Seed German Universities directly into the edu-search-service universities table.
+
+This script imports the same university data as load_university_seeds.py
+but writes directly to the PostgreSQL universities table used by the crawler.
+"""
+
+import psycopg2
+import os
+import sys
+
+# Add the backend scripts path to import university data
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), '../../backend/scripts'))
+from load_university_seeds import (
+    UNIVERSITAETEN, FACHHOCHSCHULEN, PAEDAGOGISCHE_HOCHSCHULEN,
+    KUNSTHOCHSCHULEN, PRIVATE_HOCHSCHULEN
+)
+
+# Database connection from environment or defaults
+DATABASE_URL = os.environ.get(
+    'DATABASE_URL',
+    'postgresql://breakpilot:breakpilot@localhost:5432/breakpilot_db'
+)
+
+
+def get_uni_type(original_type: str) -> str:
+    """Map the type from seed data to database uni_type."""
+    type_map = {
+        'UNI': 'UNI',
+        'FH': 'HAW',  # Fachhochschule -> HAW (Hochschule für Angewandte Wissenschaften)
+        'PH': 'PH',   # Pädagogische Hochschule
+        'KUNST': 'KUNST',
+        'PRIVATE': 'PRIVATE',
+    }
+    return type_map.get(original_type, 'UNI')
+
+
+def seed_universities():
+    """Load all universities into the database."""
+    # Collect all universities with their types
+    all_unis = []
+
+    for uni in UNIVERSITAETEN:
+        all_unis.append({**uni, 'uni_type': 'UNI'})
+
+    for uni in FACHHOCHSCHULEN:
+        all_unis.append({**uni, 'uni_type': 'HAW'})
+
+    for uni in PAEDAGOGISCHE_HOCHSCHULEN:
+        all_unis.append({**uni, 'uni_type': 'PH'})
+
+    for uni in KUNSTHOCHSCHULEN:
+        all_unis.append({**uni, 'uni_type': 'KUNST'})
+
+    for uni in PRIVATE_HOCHSCHULEN:
+        all_unis.append({**uni, 'uni_type': 'PRIVATE'})
+
+    print(f"Total universities to seed: {len(all_unis)}")
+    print(f"  - Universitäten: {len(UNIVERSITAETEN)}")
+    print(f"  - Fachhochschulen: {len(FACHHOCHSCHULEN)}")
+    print(f"  - Pädagogische Hochschulen: {len(PAEDAGOGISCHE_HOCHSCHULEN)}")
+    print(f"  - Kunst-/Musikhochschulen: {len(KUNSTHOCHSCHULEN)}")
+    print(f"  - Private Hochschulen: {len(PRIVATE_HOCHSCHULEN)}")
+
+    try:
+        conn = psycopg2.connect(DATABASE_URL)
+        cur = conn.cursor()
+
+        inserted = 0
+        skipped = 0
+        errors = []
+
+        for uni in all_unis:
+            try:
+                # Generate a short name from the full name
+                name = uni['name']
+                short_name = None
+
+                # Try to extract common abbreviations
+                if 'KIT' in name:
+                    short_name = 'KIT'
+                elif 'TUM' in name or name == 'Technische Universität München':
+                    short_name = 'TUM'
+                elif 'LMU' in name or 'Ludwig-Maximilians' in name:
+                    short_name = 'LMU'
+                elif 'RWTH' in name:
+                    short_name = 'RWTH'
+                elif 'FAU' in name or 'Friedrich-Alexander' in name:
+                    short_name = 'FAU'
+                elif name.startswith('Universität '):
+                    short_name = 'Uni ' + name.replace('Universität ', '')[:15]
+                elif name.startswith('Technische Universität '):
+                    short_name = 'TU ' + name.replace('Technische Universität ', '')[:12]
+                elif name.startswith('Hochschule '):
+                    short_name = 'HS ' + name.replace('Hochschule ', '')[:15]
+
+                cur.execute("""
+                    INSERT INTO universities (name, short_name, url, state, uni_type)
+                    VALUES (%s, %s, %s, %s, %s)
+                    ON CONFLICT (url) DO NOTHING
+                    RETURNING id
+                """, (
+                    uni['name'],
+                    short_name,
+                    uni['url'],
+                    uni.get('state'),
+                    uni['uni_type']
+                ))
+
+                result = cur.fetchone()
+                if result:
+                    inserted += 1
+                else:
+                    skipped += 1
+
+            except Exception as e:
+                errors.append(f"{uni['name']}: {str(e)}")
+
+        conn.commit()
+        cur.close()
+        conn.close()
+
+        print(f"\nResults:")
+        print(f"  Inserted: {inserted}")
+        print(f"  Skipped (duplicates): {skipped}")
+
+        if errors:
+            print(f"  Errors: {len(errors)}")
+            for err in errors[:5]:
+                print(f"    - {err}")
+
+        print(f"\nDone! Total universities in database: {inserted + skipped}")
+        return True
+
+    except psycopg2.Error as e:
+        print(f"Database error: {e}")
+        return False
+
+
+if __name__ == "__main__":
+    print("=" * 60)
+    print("Seeding Universities into edu-search-service database")
+    print("=" * 60)
+
+    success = seed_universities()
+    sys.exit(0 if success else 1)
--- a/edu-search-service/scripts/vast_ai_extractor.py
+++ b/edu-search-service/scripts/vast_ai_extractor.py
@@ -0,0 +1,320 @@
+#!/usr/bin/env python3
+"""
+vast.ai Profile Extractor Script
+Dieses Skript läuft auf vast.ai und extrahiert Profildaten von Universitäts-Webseiten.
+
+Verwendung auf vast.ai:
+1. Lade dieses Skript auf deine vast.ai Instanz
+2. Installiere Abhängigkeiten: pip install requests beautifulsoup4 openai
+3. Setze Umgebungsvariablen:
+   - BREAKPILOT_API_URL=http://deine-ip:8086
+   - BREAKPILOT_API_KEY=dev-key
+   - OPENAI_API_KEY=sk-...
+4. Starte: python vast_ai_extractor.py
+"""
+
+import os
+import sys
+import json
+import time
+import logging
+import requests
+from bs4 import BeautifulSoup
+from typing import Optional, Dict, Any, List
+
+# Logging Setup
+logging.basicConfig(
+    level=logging.INFO,
+    format='%(asctime)s - %(levelname)s - %(message)s'
+)
+logger = logging.getLogger(__name__)
+
+# Configuration
+API_URL = os.environ.get('BREAKPILOT_API_URL', 'http://localhost:8086')
+API_KEY = os.environ.get('BREAKPILOT_API_KEY', 'dev-key')
+OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY', '')
+BATCH_SIZE = 10
+SLEEP_BETWEEN_REQUESTS = 1  # Sekunden zwischen Requests (respektiere rate limits)
+
+
+def fetch_pending_profiles(limit: int = 50) -> List[Dict]:
+    """Hole Profile die noch extrahiert werden müssen."""
+    try:
+        response = requests.get(
+            f"{API_URL}/api/v1/ai/extraction/pending",
+            params={"limit": limit},
+            headers={"Authorization": f"Bearer {API_KEY}"},
+            timeout=30
+        )
+        response.raise_for_status()
+        data = response.json()
+        return data.get("tasks", [])
+    except Exception as e:
+        logger.error(f"Fehler beim Abrufen der Profile: {e}")
+        return []
+
+
+def fetch_profile_page(url: str) -> Optional[str]:
+    """Lade den HTML-Inhalt einer Profilseite."""
+    try:
+        headers = {
+            'User-Agent': 'Mozilla/5.0 (compatible; BreakPilot-Crawler/1.0; +https://breakpilot.de)',
+            'Accept': 'text/html,application/xhtml+xml',
+            'Accept-Language': 'de-DE,de;q=0.9,en;q=0.8',
+        }
+        response = requests.get(url, headers=headers, timeout=30)
+        response.raise_for_status()
+        return response.text
+    except Exception as e:
+        logger.error(f"Fehler beim Laden von {url}: {e}")
+        return None
+
+
+def extract_with_beautifulsoup(html: str, url: str) -> Dict[str, Any]:
+    """Extrahiere Basis-Informationen mit BeautifulSoup (ohne AI)."""
+    soup = BeautifulSoup(html, 'html.parser')
+    data = {}
+
+    # Email suchen
+    email_links = soup.find_all('a', href=lambda x: x and x.startswith('mailto:'))
+    if email_links:
+        email = email_links[0]['href'].replace('mailto:', '').split('?')[0]
+        data['email'] = email
+
+    # Telefon suchen
+    phone_links = soup.find_all('a', href=lambda x: x and x.startswith('tel:'))
+    if phone_links:
+        data['phone'] = phone_links[0]['href'].replace('tel:', '')
+
+    # ORCID suchen
+    orcid_links = soup.find_all('a', href=lambda x: x and 'orcid.org' in x)
+    if orcid_links:
+        orcid = orcid_links[0]['href']
+        # Extrahiere ORCID ID
+        if '/' in orcid:
+            data['orcid'] = orcid.split('/')[-1]
+
+    # Google Scholar suchen
+    scholar_links = soup.find_all('a', href=lambda x: x and 'scholar.google' in x)
+    if scholar_links:
+        href = scholar_links[0]['href']
+        if 'user=' in href:
+            data['google_scholar_id'] = href.split('user=')[1].split('&')[0]
+
+    # ResearchGate suchen
+    rg_links = soup.find_all('a', href=lambda x: x and 'researchgate.net' in x)
+    if rg_links:
+        data['researchgate_url'] = rg_links[0]['href']
+
+    # LinkedIn suchen
+    linkedin_links = soup.find_all('a', href=lambda x: x and 'linkedin.com' in x)
+    if linkedin_links:
+        data['linkedin_url'] = linkedin_links[0]['href']
+
+    # Institut/Abteilung Links sammeln (für Hierarchie-Erkennung)
+    base_domain = '/'.join(url.split('/')[:3])
+    department_links = []
+    for link in soup.find_all('a', href=True):
+        href = link['href']
+        text = link.get_text(strip=True)
+        # Suche nach Links die auf Institute/Fakultäten hindeuten
+        if any(kw in text.lower() for kw in ['institut', 'fakultät', 'fachbereich', 'abteilung', 'lehrstuhl']):
+            if href.startswith('/'):
+                href = base_domain + href
+            if href.startswith('http'):
+                department_links.append({'url': href, 'name': text})
+
+    if department_links:
+        # Nimm den ersten gefundenen Department-Link
+        data['department_url'] = department_links[0]['url']
+        data['department_name'] = department_links[0]['name']
+
+    return data
+
+
+def extract_with_ai(html: str, url: str, full_name: str) -> Dict[str, Any]:
+    """Extrahiere strukturierte Daten mit OpenAI GPT."""
+    if not OPENAI_API_KEY:
+        logger.warning("Kein OPENAI_API_KEY gesetzt - nutze nur BeautifulSoup")
+        return extract_with_beautifulsoup(html, url)
+
+    try:
+        import openai
+        client = openai.OpenAI(api_key=OPENAI_API_KEY)
+
+        # Reduziere HTML auf relevanten Text
+        soup = BeautifulSoup(html, 'html.parser')
+
+        # Entferne Scripts, Styles, etc.
+        for tag in soup(['script', 'style', 'nav', 'footer', 'header']):
+            tag.decompose()
+
+        # Extrahiere Text
+        text = soup.get_text(separator='\n', strip=True)
+        # Limitiere auf 8000 Zeichen für API
+        text = text[:8000]
+
+        prompt = f"""Analysiere diese Universitäts-Profilseite für {full_name} und extrahiere folgende Informationen im JSON-Format:
+
+{{
+  "email": "email@uni.de oder null",
+  "phone": "Telefonnummer oder null",
+  "office": "Raum/Büro oder null",
+  "position": "Position/Titel (z.B. Wissenschaftlicher Mitarbeiter, Professorin) oder null",
+  "department_name": "Name des Instituts/der Abteilung oder null",
+  "research_interests": ["Liste", "der", "Forschungsthemen"] oder [],
+  "teaching_topics": ["Liste", "der", "Lehrveranstaltungen/Fächer"] oder [],
+  "supervisor_name": "Name des Vorgesetzten/Lehrstuhlinhabers falls erkennbar oder null"
+}}
+
+Profilseite von {url}:
+
+{text}
+
+Antworte NUR mit dem JSON-Objekt, keine Erklärungen."""
+
+        response = client.chat.completions.create(
+            model="gpt-4o-mini",  # Kostengünstig und schnell
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0.1,
+            max_tokens=500
+        )
+
+        result_text = response.choices[0].message.content.strip()
+
+        # Parse JSON (entferne eventuelle Markdown-Blöcke)
+        if result_text.startswith('```'):
+            result_text = result_text.split('```')[1]
+            if result_text.startswith('json'):
+                result_text = result_text[4:]
+
+        ai_data = json.loads(result_text)
+
+        # Kombiniere mit BeautifulSoup-Ergebnissen (für Links wie ORCID)
+        bs_data = extract_with_beautifulsoup(html, url)
+
+        # AI-Daten haben Priorität, aber BS-Daten für spezifische Links
+        for key in ['orcid', 'google_scholar_id', 'researchgate_url', 'linkedin_url']:
+            if key in bs_data and bs_data[key]:
+                ai_data[key] = bs_data[key]
+
+        return ai_data
+
+    except Exception as e:
+        logger.error(f"AI-Extraktion fehlgeschlagen: {e}")
+        return extract_with_beautifulsoup(html, url)
+
+
+def submit_extracted_data(staff_id: str, data: Dict[str, Any]) -> bool:
+    """Sende extrahierte Daten zurück an BreakPilot."""
+    try:
+        payload = {"staff_id": staff_id, **data}
+
+        # Entferne None-Werte
+        payload = {k: v for k, v in payload.items() if v is not None}
+
+        response = requests.post(
+            f"{API_URL}/api/v1/ai/extraction/submit",
+            json=payload,
+            headers={
+                "Authorization": f"Bearer {API_KEY}",
+                "Content-Type": "application/json"
+            },
+            timeout=30
+        )
+        response.raise_for_status()
+        return True
+    except Exception as e:
+        logger.error(f"Fehler beim Senden der Daten für {staff_id}: {e}")
+        return False
+
+
+def process_profiles():
+    """Hauptschleife: Hole Profile, extrahiere Daten, sende zurück."""
+    logger.info(f"Starte Extraktion - API: {API_URL}")
+
+    processed = 0
+    errors = 0
+
+    while True:
+        # Hole neue Profile
+        profiles = fetch_pending_profiles(limit=BATCH_SIZE)
+
+        if not profiles:
+            logger.info("Keine weiteren Profile zum Verarbeiten. Warte 60 Sekunden...")
+            time.sleep(60)
+            continue
+
+        logger.info(f"Verarbeite {len(profiles)} Profile...")
+
+        for profile in profiles:
+            staff_id = profile['staff_id']
+            url = profile['profile_url']
+            full_name = profile.get('full_name', 'Unbekannt')
+
+            logger.info(f"Verarbeite: {full_name} - {url}")
+
+            # Lade Profilseite
+            html = fetch_profile_page(url)
+            if not html:
+                errors += 1
+                continue
+
+            # Extrahiere Daten
+            extracted = extract_with_ai(html, url, full_name)
+
+            if extracted:
+                # Sende zurück
+                if submit_extracted_data(staff_id, extracted):
+                    processed += 1
+                    logger.info(f"Erfolgreich: {full_name} - Email: {extracted.get('email', 'N/A')}")
+                else:
+                    errors += 1
+            else:
+                errors += 1
+
+            # Rate limiting
+            time.sleep(SLEEP_BETWEEN_REQUESTS)
+
+        logger.info(f"Batch abgeschlossen. Gesamt: {processed} erfolgreich, {errors} Fehler")
+
+
+def main():
+    """Einstiegspunkt."""
+    logger.info("=" * 60)
+    logger.info("BreakPilot vast.ai Profile Extractor")
+    logger.info("=" * 60)
+
+    # Prüfe Konfiguration
+    if not API_KEY:
+        logger.error("BREAKPILOT_API_KEY nicht gesetzt!")
+        sys.exit(1)
+
+    if not OPENAI_API_KEY:
+        logger.warning("OPENAI_API_KEY nicht gesetzt - nutze nur BeautifulSoup-Extraktion")
+
+    # Teste Verbindung
+    try:
+        response = requests.get(
+            f"{API_URL}/v1/health",
+            headers={"Authorization": f"Bearer {API_KEY}"},
+            timeout=10
+        )
+        logger.info(f"API-Verbindung OK: {response.status_code}")
+    except Exception as e:
+        logger.error(f"Kann API nicht erreichen: {e}")
+        logger.error(f"Stelle sicher dass {API_URL} erreichbar ist!")
+        sys.exit(1)
+
+    # Starte Verarbeitung
+    try:
+        process_profiles()
+    except KeyboardInterrupt:
+        logger.info("Beendet durch Benutzer")
+    except Exception as e:
+        logger.error(f"Unerwarteter Fehler: {e}")
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    main()
--- a/edu-search-service/seeds/de_federal.txt
+++ b/edu-search-service/seeds/de_federal.txt
@@ -0,0 +1,25 @@
+# Bundesebene - Offizielle Bildungsquellen
+# Format: URL [optional: max_depth]
+
+# Kultusministerkonferenz
+https://www.kmk.org
+https://www.kmk.org/themen/qualitaetssicherung-in-schulen/bildungsstandards.html
+https://www.kmk.org/themen/allgemeinbildende-schulen/lehrplaene-curricula.html
+
+# Deutscher Bildungsserver (DIPF)
+https://www.bildungsserver.de
+https://www.bildungsserver.de/Lehrplaene-400.html
+https://www.bildungsserver.de/Unterrichtsmaterial-389.html
+https://www.bildungsserver.de/Schulsystem-532.html
+
+# Bundeszentrale für politische Bildung
+https://www.bpb.de/lernen
+https://www.bpb.de/shop/materialien/arbeitsblatt
+https://www.bpb.de/themen
+
+# BMBF
+https://www.bmbf.de/bmbf/de/bildung/bildung_node.html
+
+# IQB - Institut zur Qualitätsentwicklung im Bildungswesen
+https://www.iqb.hu-berlin.de/bista
+https://www.iqb.hu-berlin.de/vera
--- a/edu-search-service/seeds/de_laender.txt
+++ b/edu-search-service/seeds/de_laender.txt
@@ -0,0 +1,74 @@
+# Bundesländer - Kultusministerien und Bildungsserver
+# Format: URL # Bundesland Kategorie
+
+# Baden-Württemberg
+https://km-bw.de # BW Ministerium
+https://www.bildungsplaene-bw.de # BW Lehrpläne
+https://www.schule-bw.de # BW Bildungsserver
+https://lehrerfortbildung-bw.de # BW Fortbildung
+
+# Bayern
+https://www.km.bayern.de # BY Ministerium
+https://www.lehrplanplus.bayern.de # BY Lehrpläne
+https://www.isb.bayern.de # BY ISB
+
+# Berlin
+https://www.berlin.de/sen/bildung # BE Senat
+https://bildungsserver.berlin-brandenburg.de # BE BB Bildungsserver
+
+# Brandenburg
+https://mbjs.brandenburg.de # BB Ministerium
+https://bildungsserver.berlin-brandenburg.de # BB Bildungsserver
+
+# Bremen
+https://www.bildung.bremen.de # HB Bildung
+
+# Hamburg
+https://www.hamburg.de/bsb # HH Behörde
+https://bildungsserver.hamburg.de # HH Bildungsserver
+
+# Hessen
+https://kultusministerium.hessen.de # HE Ministerium
+https://www.schulportal.hessen.de # HE Schulportal
+
+# Mecklenburg-Vorpommern
+https://www.bildung-mv.de # MV Bildung
+https://www.bildungsserver-mv.de # MV Bildungsserver
+
+# Niedersachsen
+https://www.mk.niedersachsen.de # NI Ministerium
+https://www.nibis.de # NI Bildungsserver
+https://cuvo.nibis.de # NI Curricula
+
+# Nordrhein-Westfalen
+https://www.schulministerium.nrw.de # NW Ministerium
+https://www.schulentwicklung.nrw.de # NW Entwicklung
+https://www.standardsicherung.nrw.de # NW Lehrpläne
+https://www.learnline.nrw.de # NW Bildungsserver
+
+# Rheinland-Pfalz
+https://bm.rlp.de # RP Ministerium
+https://lehrplaene.bildung-rp.de # RP Lehrpläne
+https://schuleonline.bildung-rp.de # RP Bildungsserver
+
+# Saarland
+https://www.saarland.de/mbk # SL Ministerium
+https://www.lpm.uni-sb.de # SL LPM
+
+# Sachsen
+https://www.schule.sachsen.de # SN Schule
+https://www.sachsen-macht-schule.de # SN Portal
+https://www.schulportal.sachsen.de # SN Schulportal
+
+# Sachsen-Anhalt
+https://mb.sachsen-anhalt.de # ST Ministerium
+https://www.bildung-lsa.de # ST Bildungsserver
+
+# Schleswig-Holstein
+https://www.schleswig-holstein.de/DE/landesregierung/ministerien-behoerden/III # SH Ministerium
+https://lehrplan.sh # SH Lehrpläne
+https://fachportal.lernnetz.de # SH Fachportal
+
+# Thüringen
+https://bildung.thueringen.de # TH Bildung
+https://www.schulportal-thueringen.de # TH Schulportal
--- a/edu-search-service/seeds/de_portals.txt
+++ b/edu-search-service/seeds/de_portals.txt
@@ -0,0 +1,20 @@
+# Lehrerportale und Materialsammlungen
+# Format: URL # Kategorie
+
+# Große Lehrerportale
+https://www.lehrer-online.de # Unterrichtsmaterial
+https://www.4teachers.de # Lehrerportal
+https://www.schulportal.de # Schulportal
+
+# Open Educational Resources
+https://www.oer-info.de # OER Portal
+https://www.zum.de # ZUM Zentrale
+https://wiki.zum.de # ZUM Wiki
+
+# Fachspezifische Portale
+https://unterricht.schule # Unterricht
+https://www.lernen-mit-spass.ch # Lernhilfen
+
+# Universitäten mit Lehrmaterialien
+https://www.uni-due.de/ludi # Lehrerbildung
+https://www.tu-dortmund.de/uni/Uni/Fakultaeten/FK12 # Fachdidaktik
--- a/edu-search-service/seeds/denylist.txt
+++ b/edu-search-service/seeds/denylist.txt
@@ -0,0 +1,23 @@
+# Denylist - Diese Domains werden niemals gecrawlt
+# Format: domain
+
+# Werbung/Tracking
+doubleclick.net
+googleadservices.com
+facebook.com
+twitter.com
+
+# Social Media
+instagram.com
+tiktok.com
+youtube.com
+
+# E-Commerce
+amazon.de
+ebay.de
+
+# Paywalled Content
+spiegel.de
+zeit.de
+sueddeutsche.de
+faz.net
--- a/geo-service/.env.example
+++ b/geo-service/.env.example
@@ -1,81 +0,0 @@
-# GeoEdu Service Environment Configuration
-# Copy this file to .env and adjust values as needed
-
-# ===========================================
-# Service Configuration
-# ===========================================
-PORT=8088
-ENVIRONMENT=development
-DEBUG=false
-
-# ===========================================
-# JWT Authentication
-# ===========================================
-JWT_SECRET=your-super-secret-jwt-key-change-in-production
-JWT_ALGORITHM=HS256
-JWT_EXPIRATION_HOURS=24
-
-# ===========================================
-# PostgreSQL (PostGIS)
-# ===========================================
-# Note: Database must have PostGIS extension enabled
-DATABASE_URL=postgresql+asyncpg://breakpilot:breakpilot123@postgres:5432/breakpilot_db
-
-# ===========================================
-# MinIO Object Storage (S3-compatible)
-# ===========================================
-# Used for storing AOI bundles and generated assets
-MINIO_ENDPOINT=minio:9000
-MINIO_ACCESS_KEY=breakpilot
-MINIO_SECRET_KEY=breakpilot123
-MINIO_BUCKET=breakpilot-geo
-MINIO_SECURE=false
-
-# ===========================================
-# Ollama LLM (for Learning Node Generation)
-# ===========================================
-# DSGVO-compliant local LLM for generating learning content
-OLLAMA_BASE_URL=http://host.docker.internal:11434
-OLLAMA_MODEL=qwen2.5:14b
-OLLAMA_TIMEOUT=120
-
-# ===========================================
-# Data Directories (Docker Volumes)
-# ===========================================
-OSM_DATA_DIR=/app/data/osm
-DEM_DATA_DIR=/app/data/dem
-TILE_CACHE_DIR=/app/cache/tiles
-BUNDLE_DIR=/app/bundles
-
-# ===========================================
-# Tile Server Configuration
-# ===========================================
-DEFAULT_PMTILES_FILE=germany.pmtiles
-TILE_CACHE_MAX_SIZE_GB=50.0
-
-# ===========================================
-# DEM (Digital Elevation Model) Configuration
-# ===========================================
-# Copernicus DEM GLO-30 (30m resolution)
-DEM_RESOLUTION=GLO-30
-TERRAIN_TILE_SIZE=256
-
-# ===========================================
-# AOI (Area of Interest) Limits
-# ===========================================
-# DSGVO data minimization: limit area size
-MAX_AOI_SIZE_KM2=4.0
-MAX_AOI_PER_USER=10
-AOI_RETENTION_DAYS=30
-
-# ===========================================
-# Learning Nodes Configuration
-# ===========================================
-MAX_NODES_PER_AOI=20
-# Supported themes: topographie, landnutzung, orientierung, geologie, hydrologie, vegetation
-
-# ===========================================
-# CORS Configuration
-# ===========================================
-# Comma-separated list of allowed origins
-CORS_ORIGINS=http://localhost:3000,http://localhost:3001,http://localhost:8088
--- a/geo-service/Dockerfile
+++ b/geo-service/Dockerfile
@@ -1,69 +0,0 @@
-# GeoEdu Service - Self-Hosted OSM + Terrain
-# DSGVO-konform, keine externen Tile-Services
-FROM python:3.11-slim-bookworm
-
-# Build arguments
-ARG TARGETARCH
-
-# Install system dependencies for geo-processing
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    # Build essentials
-    build-essential \
-    gcc \
-    g++ \
-    # Geo libraries
-    libgdal-dev \
-    gdal-bin \
-    libgeos-dev \
-    libproj-dev \
-    # Image processing
-    libpng-dev \
-    libjpeg-dev \
-    # Network tools
-    curl \
-    wget \
-    # Clean up
-    && rm -rf /var/lib/apt/lists/*
-
-# Set GDAL environment variables
-ENV GDAL_CONFIG=/usr/bin/gdal-config
-ENV CPLUS_INCLUDE_PATH=/usr/include/gdal
-ENV C_INCLUDE_PATH=/usr/include/gdal
-
-# Create app directory
-WORKDIR /app
-
-# Create non-root user for security
-RUN groupadd -r geoservice && useradd -r -g geoservice geoservice
-
-# Create data and cache directories
-RUN mkdir -p /app/data/osm /app/data/dem /app/cache/tiles /app/bundles \
-    && chown -R geoservice:geoservice /app
-
-# Copy requirements first for better caching
-COPY requirements.txt .
-
-# Install Python dependencies
-RUN pip install --no-cache-dir -r requirements.txt
-
-# Copy application code
-COPY --chown=geoservice:geoservice . .
-
-# Create __init__.py files for Python packages
-RUN touch /app/api/__init__.py \
-    && touch /app/services/__init__.py \
-    && touch /app/models/__init__.py \
-    && touch /app/utils/__init__.py
-
-# Switch to non-root user
-USER geoservice
-
-# Expose port
-EXPOSE 8088
-
-# Health check
-HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
-    CMD curl -f http://localhost:8088/health || exit 1
-
-# Start application
-CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8088"]
--- a/geo-service/STATUS.md
+++ b/geo-service/STATUS.md
@@ -1,196 +0,0 @@
-# GeoEdu Service - Implementierungsstatus
-
-**Stand:** 2026-01-24
-**Status:** Infrastruktur komplett, Downloads ausstehend
-
---
-
-## Übersicht
-
-Der GeoEdu Service ist eine DSGVO-konforme Erdkunde-Lernplattform mit:
- Self-Hosted OpenStreetMap Tile Server für Deutschland
- 3D-Terrain aus Copernicus DEM (Open Data)
- Unity WebGL Integration für immersive Lernwelten
- Learning Nodes System für didaktische Inhalte (via Ollama LLM)
-
-**Port:** 8088
-
---
-
-## Abgeschlossene Arbeiten
-
-### Backend (geo-service/)
-
-| Datei | Status | Beschreibung |
-|-------|--------|--------------|
-| `Dockerfile` | ✅ | Python 3.11 + GDAL + Geo-Libraries |
-| `requirements.txt` | ✅ | FastAPI, asyncpg, shapely, rasterio, pmtiles |
-| `config.py` | ✅ | Pydantic Settings |
-| `main.py` | ✅ | FastAPI App mit allen Routers |
-| `api/tiles.py` | ✅ | Vector Tile Endpoints |
-| `api/terrain.py` | ✅ | DEM/Heightmap Endpoints |
-| `api/aoi.py` | ✅ | AOI Packaging Endpoints |
-| `api/learning.py` | ✅ | Learning Nodes API |
-| `services/tile_server.py` | ✅ | PMTiles Reader |
-| `services/dem_service.py` | ✅ | Copernicus DEM Handler |
-| `services/aoi_packager.py` | ✅ | Unity Bundle Generator |
-| `services/osm_extractor.py` | ✅ | OSM Feature Extraction |
-| `services/learning_generator.py` | ✅ | Ollama LLM Integration |
-| `models/aoi.py` | ✅ | AOI Pydantic Models |
-| `models/learning_node.py` | ✅ | Learning Node Models |
-| `models/attribution.py` | ✅ | License Attribution |
-| `utils/geo_utils.py` | ✅ | Koordinaten-Transformationen |
-| `utils/minio_client.py` | ✅ | S3 Client |
-| `utils/license_checker.py` | ✅ | Lizenz-Validierung |
-| `scripts/download_osm.sh` | ✅ | Geofabrik Download (NICHT ausgeführt) |
-| `scripts/download_dem.sh` | ✅ | Copernicus Download (NICHT ausgeführt) |
-| `scripts/import_osm.sh` | ✅ | osm2pgsql Import |
-| `scripts/generate_tiles.sh` | ✅ | PMTiles Generation |
-| `scripts/init_postgis.sql` | ✅ | PostGIS Schema + Tabellen |
-| `tests/test_tiles.py` | ✅ | API Tests |
-| `tests/test_aoi.py` | ✅ | AOI + Geo Utils Tests |
-| `tests/test_learning.py` | ✅ | Learning Generator Tests |
-| `.env.example` | ✅ | Environment Template |
-
-### Frontend (studio-v2/)
-
-| Datei | Status | Beschreibung |
-|-------|--------|--------------|
-| `package.json` | ✅ | maplibre-gl Dependency hinzugefügt |
-| `app/geo-lernwelt/page.tsx` | ✅ | Hauptseite mit Tabs |
-| `app/geo-lernwelt/types.ts` | ✅ | TypeScript Interfaces |
-| `components/geo-lernwelt/AOISelector.tsx` | ✅ | MapLibre Polygon-Zeichnung |
-| `components/geo-lernwelt/UnityViewer.tsx` | ✅ | Unity WebGL Viewer |
-| `lib/geo-lernwelt/GeoContext.tsx` | ✅ | React Context |
-| `lib/geo-lernwelt/mapStyles.ts` | ✅ | MapLibre Styles |
-| `lib/geo-lernwelt/unityBridge.ts` | ✅ | Unity Communication |
-
-### Docker-Konfiguration
-
-| Änderung | Status | Beschreibung |
-|----------|--------|--------------|
-| `docker-compose.yml` | ✅ | geo-service hinzugefügt |
-| PostgreSQL Image | ✅ | Geändert zu `postgis/postgis:16-3.4-alpine` |
-| PostGIS Init Script | ✅ | Auto-Setup bei Container-Start |
-| Volumes | ✅ | geo_osm_data, geo_dem_data, geo_tile_cache, geo_aoi_bundles |
-
---
-
-## Ausstehende Arbeiten
-
-### Downloads (NICHT automatisch starten!)
-
-| Download | Größe | Script | Dauer (~100 Mbit/s) |
-|----------|-------|--------|---------------------|
-| Germany OSM PBF | 4.4 GB | `scripts/download_osm.sh` | ~6 Min |
-| Copernicus DEM | ~25 GB | `scripts/download_dem.sh` | ~35 Min |
-
-### Generierung (nach Downloads)
-
-| Prozess | Output | Dauer |
-|---------|--------|-------|
-| osm2pgsql Import | PostgreSQL DB | ~2-4 Stunden |
-| PMTiles (Z0-14) | ~200-300 GB | ~12-24 Stunden |
-| DEM Tiles | ~15 GB | ~2-3 Stunden |
-
-**Gesamtspeicherbedarf nach Setup: ~350 GB**
-
---
-
-## API Endpoints
-
-```
-# Tiles
-GET  /api/v1/tiles/{z}/{x}/{y}.pbf     → Vector Tile
-GET  /api/v1/tiles/style.json          → MapLibre Style
-GET  /api/v1/tiles/metadata            → Tile Metadata
-GET  /api/v1/tiles/bounds              → Germany Bounds
-
-# Terrain
-GET  /api/v1/terrain/{z}/{x}/{y}.png   → Heightmap (Terrain-RGB)
-GET  /api/v1/terrain/hillshade/{z}/{x}/{y}.png → Hillshade
-GET  /api/v1/terrain/elevation?lat=&lon= → Point Elevation
-GET  /api/v1/terrain/metadata          → DEM Metadata
-
-# AOI (Area of Interest)
-POST /api/v1/aoi                       → Create AOI
-GET  /api/v1/aoi/{id}                  → Get Status
-GET  /api/v1/aoi/{id}/manifest.json    → Unity Manifest
-POST /api/v1/aoi/validate              → Validate Polygon
-GET  /api/v1/aoi/templates/mainau      → Mainau Demo
-
-# Learning Nodes
-POST /api/v1/learning/generate         → Generate with LLM
-GET  /api/v1/learning/templates        → Available Themes
-GET  /api/v1/learning/{aoi_id}/nodes   → Get Nodes
-GET  /api/v1/learning/statistics       → Stats
-```
-
---
-
-## Nächste Schritte
-
-1. **Container starten (ohne Downloads):**
-   ```bash
-   docker compose up -d geo-service
-   ```
-
-2. **Health Check testen:**
-   ```bash
-   curl http://localhost:8088/health
-   ```
-
-3. **Frontend testen:**
-   ```
-   http://localhost:3001/geo-lernwelt
-   ```
-
-4. **Downloads starten (wenn bereit):**
-   ```bash
-   # OSM Download (~6 Min)
-   docker exec breakpilot-pwa-geo-service /app/scripts/download_osm.sh
-
-   # DEM Download (~35 Min)
-   docker exec breakpilot-pwa-geo-service /app/scripts/download_dem.sh
-   ```
-
-5. **Tile-Generierung (nach Downloads):**
-   ```bash
-   # OSM Import (~2-4h)
-   docker exec breakpilot-pwa-geo-service /app/scripts/import_osm.sh
-
-   # PMTiles Generation (~12-24h)
-   docker exec breakpilot-pwa-geo-service /app/scripts/generate_tiles.sh
-   ```
-
---
-
-## Lizenz-Compliance
-
-### Zugelassene Quellen
- ✅ OpenStreetMap (ODbL)
- ✅ Copernicus DEM (Copernicus License)
- ✅ OpenAerialMap (CC-BY 4.0)
-
-### VERBOTEN
- ❌ Google Maps/Earth
- ❌ Bing Maps
- ❌ Apple Maps
- ❌ HERE
-
---
-
-## Wichtige Dateien
-
-| Datei | Pfad |
-|-------|------|
-| Haupt-Plan | `/Users/benjaminadmin/.claude/plans/stateful-hugging-kahan.md` |
-| Backend | `/Users/benjaminadmin/Projekte/breakpilot-pwa/geo-service/` |
-| Frontend | `/Users/benjaminadmin/Projekte/breakpilot-pwa/studio-v2/app/geo-lernwelt/` |
-| Docker Config | `/Users/benjaminadmin/Projekte/breakpilot-pwa/docker-compose.yml` |
-
---
-
-## Kontakt bei Fragen
-
-Dieser Service wurde gemäß dem Plan in `/Users/benjaminadmin/.claude/plans/stateful-hugging-kahan.md` implementiert.
--- a/geo-service/api/init.py
+++ b/geo-service/api/init.py
@@ -1,9 +0,0 @@
-"""
-GeoEdu Service API Endpoints
-"""
-from .tiles import router as tiles_router
-from .terrain import router as terrain_router
-from .aoi import router as aoi_router
-from .learning import router as learning_router
-
-__all__ = ["tiles_router", "terrain_router", "aoi_router", "learning_router"]
--- a/geo-service/api/aoi.py
+++ b/geo-service/api/aoi.py
@@ -1,304 +0,0 @@
-"""
-AOI (Area of Interest) API Endpoints
-Handles polygon selection, validation, and Unity bundle generation
-"""
-from fastapi import APIRouter, HTTPException, Path, Query, BackgroundTasks
-from fastapi.responses import JSONResponse, FileResponse
-from pydantic import BaseModel, Field
-from typing import Optional
-from datetime import datetime
-import uuid
-import structlog
-
-from config import settings
-from models.aoi import AOIRequest, AOIResponse, AOIStatus, AOIManifest
-from services.aoi_packager import AOIPackagerService
-from services.osm_extractor import OSMExtractorService
-
-logger = structlog.get_logger(__name__)
-router = APIRouter()
-
-# Initialize services
-aoi_packager = AOIPackagerService()
-osm_extractor = OSMExtractorService()
-
-
-@router.post("", response_model=AOIResponse)
-async def create_aoi(
-    request: AOIRequest,
-    background_tasks: BackgroundTasks,
-):
-    """
-    Create a new AOI (Area of Interest) for Unity 3D export.
-
-    Validates the polygon, checks size limits (max 4 km²), and queues
-    bundle generation. Returns immediately with a status URL.
-
-    The bundle will contain:
-    - Terrain heightmap
-    - OSM features (buildings, roads, water, etc.)
-    - Learning node positions
-    - Attribution information
-    """
-    # Validate polygon
-    try:
-        area_km2 = aoi_packager.calculate_area_km2(request.polygon)
-    except ValueError as e:
-        raise HTTPException(status_code=400, detail=f"Invalid polygon: {str(e)}")
-
-    # Check size limit
-    if area_km2 > settings.max_aoi_size_km2:
-        raise HTTPException(
-            status_code=400,
-            detail=f"AOI too large: {area_km2:.2f} km² exceeds maximum of {settings.max_aoi_size_km2} km²",
-        )
-
-    # Check if polygon is within Germany bounds
-    if not aoi_packager.is_within_germany(request.polygon):
-        raise HTTPException(
-            status_code=400,
-            detail="AOI must be within Germany. Bounds: [5.87°E, 47.27°N] to [15.04°E, 55.06°N]",
-        )
-
-    # Generate AOI ID
-    aoi_id = str(uuid.uuid4())
-
-    # Create AOI record
-    aoi_data = {
-        "id": aoi_id,
-        "polygon": request.polygon,
-        "theme": request.theme,
-        "quality": request.quality,
-        "area_km2": area_km2,
-        "status": AOIStatus.QUEUED,
-        "created_at": datetime.utcnow().isoformat(),
-    }
-
-    # Start background processing
-    background_tasks.add_task(
-        aoi_packager.process_aoi,
-        aoi_id=aoi_id,
-        polygon=request.polygon,
-        theme=request.theme,
-        quality=request.quality,
-    )
-
-    logger.info(
-        "AOI created",
-        aoi_id=aoi_id,
-        area_km2=area_km2,
-        theme=request.theme,
-    )
-
-    return AOIResponse(
-        aoi_id=aoi_id,
-        status=AOIStatus.QUEUED,
-        area_km2=area_km2,
-        estimated_size_mb=aoi_packager.estimate_bundle_size_mb(area_km2, request.quality),
-        message="AOI queued for processing",
-    )
-
-
-@router.get("/{aoi_id}", response_model=AOIResponse)
-async def get_aoi_status(
-    aoi_id: str = Path(..., description="AOI UUID"),
-):
-    """
-    Get the status of an AOI processing job.
-
-    Returns current status (queued, processing, completed, failed)
-    and download URLs when ready.
-    """
-    aoi_data = await aoi_packager.get_aoi_status(aoi_id)
-
-    if aoi_data is None:
-        raise HTTPException(status_code=404, detail="AOI not found")
-
-    response = AOIResponse(
-        aoi_id=aoi_id,
-        status=aoi_data["status"],
-        area_km2=aoi_data.get("area_km2", 0),
-        estimated_size_mb=aoi_data.get("estimated_size_mb", 0),
-        message=aoi_data.get("message", ""),
-    )
-
-    # Add download URLs if completed
-    if aoi_data["status"] == AOIStatus.COMPLETED:
-        response.download_url = f"/api/v1/aoi/{aoi_id}/bundle.zip"
-        response.manifest_url = f"/api/v1/aoi/{aoi_id}/manifest.json"
-        response.completed_at = aoi_data.get("completed_at")
-
-    return response
-
-
-@router.get("/{aoi_id}/manifest.json")
-async def get_aoi_manifest(
-    aoi_id: str = Path(..., description="AOI UUID"),
-):
-    """
-    Get the Unity bundle manifest for an AOI.
-
-    The manifest contains:
-    - Terrain configuration
-    - Asset list and paths
-    - Learning node positions
-    - Attribution requirements
-    """
-    manifest = await aoi_packager.get_manifest(aoi_id)
-
-    if manifest is None:
-        raise HTTPException(status_code=404, detail="Manifest not found or AOI not ready")
-
-    return JSONResponse(content=manifest)
-
-
-@router.get("/{aoi_id}/bundle.zip")
-async def download_aoi_bundle(
-    aoi_id: str = Path(..., description="AOI UUID"),
-):
-    """
-    Download the complete AOI bundle as a ZIP file.
-
-    Contains all assets needed for Unity 3D rendering:
-    - terrain.heightmap (16-bit PNG)
-    - osm_features.json (buildings, roads, etc.)
-    - learning_nodes.json (educational content positions)
-    - attribution.json (required license notices)
-    """
-    bundle_path = await aoi_packager.get_bundle_path(aoi_id)
-
-    if bundle_path is None:
-        raise HTTPException(status_code=404, detail="Bundle not found or AOI not ready")
-
-    return FileResponse(
-        path=bundle_path,
-        filename=f"geo-lernwelt-{aoi_id[:8]}.zip",
-        media_type="application/zip",
-    )
-
-
-@router.delete("/{aoi_id}")
-async def delete_aoi(
-    aoi_id: str = Path(..., description="AOI UUID"),
-):
-    """
-    Delete an AOI and its bundle.
-
-    Implements DSGVO data minimization - users can delete their data.
-    """
-    success = await aoi_packager.delete_aoi(aoi_id)
-
-    if not success:
-        raise HTTPException(status_code=404, detail="AOI not found")
-
-    logger.info("AOI deleted", aoi_id=aoi_id)
-
-    return {"message": "AOI deleted successfully", "aoi_id": aoi_id}
-
-
-@router.get("/{aoi_id}/preview")
-async def get_aoi_preview(
-    aoi_id: str = Path(..., description="AOI UUID"),
-    width: int = Query(512, ge=64, le=2048, description="Preview width"),
-    height: int = Query(512, ge=64, le=2048, description="Preview height"),
-):
-    """
-    Get a preview image of the AOI.
-
-    Returns a rendered preview showing terrain, OSM features,
-    and learning node positions.
-    """
-    preview_data = await aoi_packager.generate_preview(aoi_id, width, height)
-
-    if preview_data is None:
-        raise HTTPException(status_code=404, detail="Preview not available")
-
-    from fastapi.responses import Response
-
-    return Response(
-        content=preview_data,
-        media_type="image/png",
-    )
-
-
-@router.post("/validate")
-async def validate_aoi_polygon(
-    polygon: dict,
-):
-    """
-    Validate an AOI polygon without creating it.
-
-    Checks:
-    - Valid GeoJSON format
-    - Within Germany bounds
-    - Within size limits
-    - Not self-intersecting
-    """
-    try:
-        # Validate geometry
-        is_valid, message = aoi_packager.validate_polygon(polygon)
-
-        if not is_valid:
-            return {
-                "valid": False,
-                "error": message,
-            }
-
-        # Calculate area
-        area_km2 = aoi_packager.calculate_area_km2(polygon)
-
-        # Check bounds
-        within_germany = aoi_packager.is_within_germany(polygon)
-
-        # Check size
-        within_size_limit = area_km2 <= settings.max_aoi_size_km2
-
-        return {
-            "valid": is_valid and within_germany and within_size_limit,
-            "area_km2": round(area_km2, 3),
-            "within_germany": within_germany,
-            "within_size_limit": within_size_limit,
-            "max_size_km2": settings.max_aoi_size_km2,
-            "estimated_bundle_size_mb": aoi_packager.estimate_bundle_size_mb(area_km2, "medium"),
-        }
-
-    except Exception as e:
-        return {
-            "valid": False,
-            "error": str(e),
-        }
-
-
-@router.get("/templates/mainau")
-async def get_mainau_template():
-    """
-    Get pre-configured AOI template for Mainau island (demo location).
-
-    Mainau is a small island in Lake Constance (Bodensee) -
-    perfect for educational geography lessons.
-    """
-    return {
-        "name": "Insel Mainau",
-        "description": "Blumeninsel im Bodensee - ideal fuer Erdkunde-Unterricht",
-        "polygon": {
-            "type": "Polygon",
-            "coordinates": [
-                [
-                    [9.1875, 47.7055],
-                    [9.1975, 47.7055],
-                    [9.1975, 47.7115],
-                    [9.1875, 47.7115],
-                    [9.1875, 47.7055],
-                ]
-            ],
-        },
-        "center": [9.1925, 47.7085],
-        "area_km2": 0.45,
-        "suggested_themes": ["topographie", "vegetation", "landnutzung"],
-        "features": [
-            "Schloss und Schlosskirche",
-            "Botanischer Garten",
-            "Bodensee-Ufer",
-            "Waldgebiete",
-        ],
-    }
--- a/geo-service/api/learning.py
+++ b/geo-service/api/learning.py
@@ -1,289 +0,0 @@
-"""
-Learning Nodes API Endpoints
-Generates and manages educational content for AOI regions
-"""
-from fastapi import APIRouter, HTTPException, Path, Query, BackgroundTasks
-from fastapi.responses import JSONResponse
-from pydantic import BaseModel, Field
-from typing import Optional
-import structlog
-
-from config import settings
-from models.learning_node import LearningNode, LearningNodeRequest, LearningTheme
-from services.learning_generator import LearningGeneratorService
-
-logger = structlog.get_logger(__name__)
-router = APIRouter()
-
-# Initialize learning generator service
-learning_service = LearningGeneratorService()
-
-
-class GenerateNodesRequest(BaseModel):
-    """Request model for generating learning nodes."""
-
-    aoi_id: str = Field(..., description="AOI UUID to generate nodes for")
-    theme: LearningTheme = Field(LearningTheme.TOPOGRAPHIE, description="Learning theme")
-    difficulty: str = Field("mittel", pattern="^(leicht|mittel|schwer)$", description="Difficulty level")
-    node_count: int = Field(5, ge=1, le=20, description="Number of nodes to generate")
-    grade_level: Optional[str] = Field(None, description="Target grade level (e.g., '5-6', '7-8')")
-    language: str = Field("de", description="Content language")
-
-
-class GenerateNodesResponse(BaseModel):
-    """Response model for generated learning nodes."""
-
-    aoi_id: str
-    theme: str
-    nodes: list[LearningNode]
-    total_count: int
-    generation_model: str
-
-
-@router.post("/generate", response_model=GenerateNodesResponse)
-async def generate_learning_nodes(
-    request: GenerateNodesRequest,
-):
-    """
-    Generate learning nodes for an AOI using LLM.
-
-    Uses Ollama with the configured model to generate educational content
-    based on the AOI's geographic features and selected theme.
-
-    Themes:
-    - topographie: Landforms, elevation, terrain features
-    - landnutzung: Land use, settlement patterns, agriculture
-    - orientierung: Navigation, compass, map reading
-    - geologie: Rock types, geological formations
-    - hydrologie: Water features, drainage, watersheds
-    - vegetation: Plant communities, climate zones
-    """
-    try:
-        nodes = await learning_service.generate_nodes(
-            aoi_id=request.aoi_id,
-            theme=request.theme,
-            difficulty=request.difficulty,
-            node_count=request.node_count,
-            grade_level=request.grade_level,
-            language=request.language,
-        )
-
-        logger.info(
-            "Learning nodes generated",
-            aoi_id=request.aoi_id,
-            theme=request.theme.value,
-            count=len(nodes),
-        )
-
-        return GenerateNodesResponse(
-            aoi_id=request.aoi_id,
-            theme=request.theme.value,
-            nodes=nodes,
-            total_count=len(nodes),
-            generation_model=settings.ollama_model,
-        )
-
-    except FileNotFoundError:
-        raise HTTPException(
-            status_code=404,
-            detail="AOI not found. Please create an AOI first.",
-        )
-    except ConnectionError:
-        raise HTTPException(
-            status_code=503,
-            detail="LLM service (Ollama) not available. Please check if Ollama is running.",
-        )
-    except Exception as e:
-        logger.error("Error generating learning nodes", error=str(e))
-        raise HTTPException(status_code=500, detail="Error generating learning nodes")
-
-
-@router.get("/templates")
-async def get_learning_templates():
-    """
-    Get available learning theme templates.
-
-    Returns theme definitions with descriptions, suitable grade levels,
-    and example questions.
-    """
-    return {
-        "themes": [
-            {
-                "id": "topographie",
-                "name": "Topographie",
-                "description": "Landschaftsformen, Höhen und Geländemerkmale",
-                "icon": "mountain",
-                "grade_levels": ["5-6", "7-8", "9-10"],
-                "example_questions": [
-                    "Welche Höhe hat der höchste Punkt in diesem Gebiet?",
-                    "Beschreibe die Hangneigung im nördlichen Bereich.",
-                    "Wo befinden sich Täler und wo Bergrücken?",
-                ],
-                "keywords": ["Höhe", "Hang", "Tal", "Berg", "Hügel", "Ebene"],
-            },
-            {
-                "id": "landnutzung",
-                "name": "Landnutzung",
-                "description": "Siedlungen, Landwirtschaft und Flächennutzung",
-                "icon": "home",
-                "grade_levels": ["5-6", "7-8", "9-10", "11-12"],
-                "example_questions": [
-                    "Welche Arten von Gebäuden sind in diesem Bereich zu finden?",
-                    "Wie viel Prozent der Fläche wird landwirtschaftlich genutzt?",
-                    "Wo verläuft die Grenze zwischen Siedlung und Naturraum?",
-                ],
-                "keywords": ["Siedlung", "Acker", "Wald", "Industrie", "Straße"],
-            },
-            {
-                "id": "orientierung",
-                "name": "Orientierung",
-                "description": "Kartenlesen, Kompass und Navigation",
-                "icon": "compass",
-                "grade_levels": ["5-6", "7-8"],
-                "example_questions": [
-                    "In welcher Himmelsrichtung liegt der See?",
-                    "Wie lang ist der Weg von A nach B?",
-                    "Beschreibe den Weg vom Parkplatz zum Aussichtspunkt.",
-                ],
-                "keywords": ["Norden", "Süden", "Entfernung", "Maßstab", "Legende"],
-            },
-            {
-                "id": "geologie",
-                "name": "Geologie",
-                "description": "Gesteinsarten und geologische Formationen",
-                "icon": "layers",
-                "grade_levels": ["7-8", "9-10", "11-12"],
-                "example_questions": [
-                    "Welches Gestein dominiert in diesem Gebiet?",
-                    "Wie sind die Felsformationen entstanden?",
-                    "Welche Rolle spielt die Geologie für die Landschaft?",
-                ],
-                "keywords": ["Gestein", "Formation", "Erosion", "Schicht", "Mineral"],
-            },
-            {
-                "id": "hydrologie",
-                "name": "Hydrologie",
-                "description": "Gewässer, Einzugsgebiete und Wasserkreislauf",
-                "icon": "droplet",
-                "grade_levels": ["5-6", "7-8", "9-10"],
-                "example_questions": [
-                    "Wohin fließt das Wasser in diesem Gebiet?",
-                    "Welche Gewässerarten sind vorhanden?",
-                    "Wo könnte sich bei Starkregen Wasser sammeln?",
-                ],
-                "keywords": ["Fluss", "See", "Bach", "Einzugsgebiet", "Quelle"],
-            },
-            {
-                "id": "vegetation",
-                "name": "Vegetation",
-                "description": "Pflanzengemeinschaften und Klimazonen",
-                "icon": "tree",
-                "grade_levels": ["5-6", "7-8", "9-10", "11-12"],
-                "example_questions": [
-                    "Welche Waldtypen sind in diesem Gebiet zu finden?",
-                    "Wie beeinflusst die Höhenlage die Vegetation?",
-                    "Welche Pflanzen würdest du hier erwarten?",
-                ],
-                "keywords": ["Wald", "Wiese", "Laubbaum", "Nadelbaum", "Höhenstufe"],
-            },
-        ],
-        "difficulties": [
-            {"id": "leicht", "name": "Leicht", "description": "Grundlegende Beobachtungen"},
-            {"id": "mittel", "name": "Mittel", "description": "Verknüpfung von Zusammenhängen"},
-            {"id": "schwer", "name": "Schwer", "description": "Analyse und Transfer"},
-        ],
-        "supported_languages": ["de", "en"],
-    }
-
-
-@router.get("/{aoi_id}/nodes")
-async def get_aoi_learning_nodes(
-    aoi_id: str = Path(..., description="AOI UUID"),
-    theme: Optional[LearningTheme] = Query(None, description="Filter by theme"),
-):
-    """
-    Get all learning nodes for an AOI.
-
-    Returns previously generated nodes, optionally filtered by theme.
-    """
-    nodes = await learning_service.get_nodes_for_aoi(aoi_id, theme)
-
-    if nodes is None:
-        raise HTTPException(status_code=404, detail="AOI not found")
-
-    return {
-        "aoi_id": aoi_id,
-        "nodes": nodes,
-        "total_count": len(nodes),
-    }
-
-
-@router.put("/{aoi_id}/nodes/{node_id}")
-async def update_learning_node(
-    aoi_id: str = Path(..., description="AOI UUID"),
-    node_id: str = Path(..., description="Node UUID"),
-    node_update: LearningNode,
-):
-    """
-    Update a learning node (teacher review/edit).
-
-    Allows teachers to modify AI-generated content before use.
-    """
-    success = await learning_service.update_node(aoi_id, node_id, node_update)
-
-    if not success:
-        raise HTTPException(status_code=404, detail="Node not found")
-
-    logger.info("Learning node updated", aoi_id=aoi_id, node_id=node_id)
-
-    return {"message": "Node updated successfully", "node_id": node_id}
-
-
-@router.delete("/{aoi_id}/nodes/{node_id}")
-async def delete_learning_node(
-    aoi_id: str = Path(..., description="AOI UUID"),
-    node_id: str = Path(..., description="Node UUID"),
-):
-    """
-    Delete a learning node.
-    """
-    success = await learning_service.delete_node(aoi_id, node_id)
-
-    if not success:
-        raise HTTPException(status_code=404, detail="Node not found")
-
-    return {"message": "Node deleted successfully", "node_id": node_id}
-
-
-@router.post("/{aoi_id}/nodes/{node_id}/approve")
-async def approve_learning_node(
-    aoi_id: str = Path(..., description="AOI UUID"),
-    node_id: str = Path(..., description="Node UUID"),
-):
-    """
-    Approve a learning node for student use.
-
-    Only approved nodes will be visible to students.
-    """
-    success = await learning_service.approve_node(aoi_id, node_id)
-
-    if not success:
-        raise HTTPException(status_code=404, detail="Node not found")
-
-    return {"message": "Node approved", "node_id": node_id}
-
-
-@router.get("/statistics")
-async def get_learning_statistics():
-    """
-    Get statistics about learning node usage.
-    """
-    stats = await learning_service.get_statistics()
-
-    return {
-        "total_nodes_generated": stats.get("total_nodes", 0),
-        "nodes_by_theme": stats.get("by_theme", {}),
-        "nodes_by_difficulty": stats.get("by_difficulty", {}),
-        "average_nodes_per_aoi": stats.get("avg_per_aoi", 0),
-        "most_popular_theme": stats.get("popular_theme", "topographie"),
-    }
--- a/geo-service/api/terrain.py
+++ b/geo-service/api/terrain.py
@@ -1,230 +0,0 @@
-"""
-Terrain/DEM API Endpoints
-Serves heightmap tiles from Copernicus DEM data
-"""
-from fastapi import APIRouter, HTTPException, Path, Query, Response
-from fastapi.responses import JSONResponse
-import structlog
-
-from config import settings
-from services.dem_service import DEMService
-
-logger = structlog.get_logger(__name__)
-router = APIRouter()
-
-# Initialize DEM service
-dem_service = DEMService()
-
-
-@router.get("/{z}/{x}/{y}.png", response_class=Response)
-async def get_heightmap_tile(
-    z: int = Path(..., ge=0, le=14, description="Zoom level"),
-    x: int = Path(..., ge=0, description="Tile X coordinate"),
-    y: int = Path(..., ge=0, description="Tile Y coordinate"),
-):
-    """
-    Get a heightmap tile as 16-bit PNG (Mapbox Terrain-RGB encoding).
-
-    Heightmaps are generated from Copernicus DEM GLO-30 (30m resolution).
-    The encoding allows for ~0.1m precision: height = -10000 + ((R * 256 * 256 + G * 256 + B) * 0.1)
-    """
-    try:
-        tile_data = await dem_service.get_heightmap_tile(z, x, y)
-
-        if tile_data is None:
-            return Response(status_code=204)  # No terrain data for this tile
-
-        return Response(
-            content=tile_data,
-            media_type="image/png",
-            headers={
-                "Cache-Control": "public, max-age=604800",  # 7 days cache
-                "X-Tile-Source": "copernicus-dem",
-            },
-        )
-
-    except FileNotFoundError:
-        logger.warning("DEM data not found", z=z, x=x, y=y)
-        raise HTTPException(
-            status_code=503,
-            detail="DEM data not available. Please download Copernicus DEM first.",
-        )
-    except Exception as e:
-        logger.error("Error serving heightmap tile", z=z, x=x, y=y, error=str(e))
-        raise HTTPException(status_code=500, detail="Error serving heightmap tile")
-
-
-@router.get("/hillshade/{z}/{x}/{y}.png", response_class=Response)
-async def get_hillshade_tile(
-    z: int = Path(..., ge=0, le=14, description="Zoom level"),
-    x: int = Path(..., ge=0, description="Tile X coordinate"),
-    y: int = Path(..., ge=0, description="Tile Y coordinate"),
-    azimuth: float = Query(315, ge=0, le=360, description="Light azimuth in degrees"),
-    altitude: float = Query(45, ge=0, le=90, description="Light altitude in degrees"),
-):
-    """
-    Get a hillshade tile for terrain visualization.
-
-    Hillshade is rendered from DEM with configurable light direction.
-    Default light comes from northwest (azimuth=315) at 45° altitude.
-    """
-    try:
-        tile_data = await dem_service.get_hillshade_tile(z, x, y, azimuth, altitude)
-
-        if tile_data is None:
-            return Response(status_code=204)
-
-        return Response(
-            content=tile_data,
-            media_type="image/png",
-            headers={
-                "Cache-Control": "public, max-age=604800",
-                "X-Hillshade-Azimuth": str(azimuth),
-                "X-Hillshade-Altitude": str(altitude),
-            },
-        )
-
-    except FileNotFoundError:
-        raise HTTPException(status_code=503, detail="DEM data not available")
-    except Exception as e:
-        logger.error("Error serving hillshade tile", z=z, x=x, y=y, error=str(e))
-        raise HTTPException(status_code=500, detail="Error serving hillshade tile")
-
-
-@router.get("/contours/{z}/{x}/{y}.pbf", response_class=Response)
-async def get_contour_tile(
-    z: int = Path(..., ge=0, le=14, description="Zoom level"),
-    x: int = Path(..., ge=0, description="Tile X coordinate"),
-    y: int = Path(..., ge=0, description="Tile Y coordinate"),
-    interval: int = Query(20, ge=5, le=100, description="Contour interval in meters"),
-):
-    """
-    Get contour lines as vector tile.
-
-    Contours are generated from DEM at the specified interval.
-    Useful for topographic map overlays.
-    """
-    try:
-        tile_data = await dem_service.get_contour_tile(z, x, y, interval)
-
-        if tile_data is None:
-            return Response(status_code=204)
-
-        return Response(
-            content=tile_data,
-            media_type="application/x-protobuf",
-            headers={
-                "Content-Encoding": "gzip",
-                "Cache-Control": "public, max-age=604800",
-                "X-Contour-Interval": str(interval),
-            },
-        )
-
-    except FileNotFoundError:
-        raise HTTPException(status_code=503, detail="DEM data not available")
-    except Exception as e:
-        logger.error("Error serving contour tile", z=z, x=x, y=y, error=str(e))
-        raise HTTPException(status_code=500, detail="Error serving contour tile")
-
-
-@router.get("/elevation")
-async def get_elevation_at_point(
-    lat: float = Query(..., ge=47.27, le=55.06, description="Latitude"),
-    lon: float = Query(..., ge=5.87, le=15.04, description="Longitude"),
-):
-    """
-    Get elevation at a specific point.
-
-    Returns elevation in meters from Copernicus DEM.
-    Only works within Germany bounds.
-    """
-    try:
-        elevation = await dem_service.get_elevation(lat, lon)
-
-        if elevation is None:
-            raise HTTPException(
-                status_code=404,
-                detail="No elevation data available for this location",
-            )
-
-        return {
-            "latitude": lat,
-            "longitude": lon,
-            "elevation_m": round(elevation, 1),
-            "source": "Copernicus DEM GLO-30",
-            "resolution_m": 30,
-        }
-
-    except FileNotFoundError:
-        raise HTTPException(status_code=503, detail="DEM data not available")
-    except Exception as e:
-        logger.error("Error getting elevation", lat=lat, lon=lon, error=str(e))
-        raise HTTPException(status_code=500, detail="Error getting elevation")
-
-
-@router.post("/elevation/profile")
-async def get_elevation_profile(
-    coordinates: list[list[float]],
-    samples: int = Query(100, ge=10, le=1000, description="Number of sample points"),
-):
-    """
-    Get elevation profile along a path.
-
-    Takes a list of [lon, lat] coordinates and returns elevations sampled along the path.
-    Useful for hiking/cycling route profiles.
-    """
-    if len(coordinates) < 2:
-        raise HTTPException(status_code=400, detail="At least 2 coordinates required")
-
-    if len(coordinates) > 100:
-        raise HTTPException(status_code=400, detail="Maximum 100 coordinates allowed")
-
-    try:
-        profile = await dem_service.get_elevation_profile(coordinates, samples)
-
-        return {
-            "profile": profile,
-            "statistics": {
-                "min_elevation_m": min(p["elevation_m"] for p in profile if p["elevation_m"]),
-                "max_elevation_m": max(p["elevation_m"] for p in profile if p["elevation_m"]),
-                "total_ascent_m": sum(
-                    max(0, profile[i + 1]["elevation_m"] - profile[i]["elevation_m"])
-                    for i in range(len(profile) - 1)
-                    if profile[i]["elevation_m"] and profile[i + 1]["elevation_m"]
-                ),
-                "total_descent_m": sum(
-                    max(0, profile[i]["elevation_m"] - profile[i + 1]["elevation_m"])
-                    for i in range(len(profile) - 1)
-                    if profile[i]["elevation_m"] and profile[i + 1]["elevation_m"]
-                ),
-            },
-            "source": "Copernicus DEM GLO-30",
-        }
-
-    except FileNotFoundError:
-        raise HTTPException(status_code=503, detail="DEM data not available")
-    except Exception as e:
-        logger.error("Error getting elevation profile", error=str(e))
-        raise HTTPException(status_code=500, detail="Error getting elevation profile")
-
-
-@router.get("/metadata")
-async def get_dem_metadata():
-    """
-    Get metadata about available DEM data.
-    """
-    metadata = await dem_service.get_metadata()
-
-    return {
-        "name": "Copernicus DEM GLO-30",
-        "description": "Global 30m Digital Elevation Model",
-        "resolution_m": 30,
-        "coverage": "Germany (Deutschland)",
-        "bounds": [5.87, 47.27, 15.04, 55.06],
-        "vertical_datum": "EGM2008",
-        "horizontal_datum": "WGS84",
-        "license": "Copernicus Data (free, attribution required)",
-        "attribution": "© Copernicus Service Information 2024",
-        "data_available": metadata.get("data_available", False),
-        "tiles_generated": metadata.get("tiles_generated", 0),
-    }
--- a/geo-service/api/tiles.py
+++ b/geo-service/api/tiles.py
@@ -1,221 +0,0 @@
-"""
-Tile Server API Endpoints
-Serves Vector Tiles from PMTiles or generates on-demand from PostGIS
-"""
-from fastapi import APIRouter, HTTPException, Path, Query, Response
-from fastapi.responses import JSONResponse
-import structlog
-
-from config import settings
-from services.tile_server import TileServerService
-
-logger = structlog.get_logger(__name__)
-router = APIRouter()
-
-# Initialize tile server service
-tile_service = TileServerService()
-
-
-@router.get("/{z}/{x}/{y}.pbf", response_class=Response)
-async def get_vector_tile(
-    z: int = Path(..., ge=0, le=22, description="Zoom level"),
-    x: int = Path(..., ge=0, description="Tile X coordinate"),
-    y: int = Path(..., ge=0, description="Tile Y coordinate"),
-):
-    """
-    Get a vector tile in Protocol Buffers format.
-
-    Returns OSM data as vector tiles suitable for MapLibre GL JS.
-    Tiles are served from pre-generated PMTiles or cached on-demand.
-    """
-    try:
-        tile_data = await tile_service.get_tile(z, x, y)
-
-        if tile_data is None:
-            # Return empty tile (204 No Content is standard for empty tiles)
-            return Response(status_code=204)
-
-        return Response(
-            content=tile_data,
-            media_type="application/x-protobuf",
-            headers={
-                "Content-Encoding": "gzip",
-                "Cache-Control": "public, max-age=86400",  # 24h cache
-                "X-Tile-Source": "pmtiles",
-            },
-        )
-
-    except FileNotFoundError:
-        logger.warning("PMTiles file not found", z=z, x=x, y=y)
-        raise HTTPException(
-            status_code=503,
-            detail="Tile data not available. Please run the data download script first.",
-        )
-    except Exception as e:
-        logger.error("Error serving tile", z=z, x=x, y=y, error=str(e))
-        raise HTTPException(status_code=500, detail="Error serving tile")
-
-
-@router.get("/style.json")
-async def get_maplibre_style(
-    base_url: str = Query(None, description="Base URL for tile server"),
-):
-    """
-    Get MapLibre GL JS style specification.
-
-    Returns a style document configured for the self-hosted tile server.
-    """
-    # Use provided base_url or construct from settings
-    if base_url is None:
-        base_url = f"http://localhost:{settings.port}"
-
-    style = {
-        "version": 8,
-        "name": "GeoEdu Germany",
-        "metadata": {
-            "description": "Self-hosted OSM tiles for DSGVO-compliant education",
-            "attribution": "© OpenStreetMap contributors",
-        },
-        "sources": {
-            "osm": {
-                "type": "vector",
-                "tiles": [f"{base_url}/api/v1/tiles/{{z}}/{{x}}/{{y}}.pbf"],
-                "minzoom": 0,
-                "maxzoom": 14,
-                "attribution": "© OpenStreetMap contributors (ODbL)",
-            },
-            "terrain": {
-                "type": "raster-dem",
-                "tiles": [f"{base_url}/api/v1/terrain/{{z}}/{{x}}/{{y}}.png"],
-                "tileSize": 256,
-                "maxzoom": 14,
-                "attribution": "© Copernicus DEM GLO-30",
-            },
-        },
-        "sprite": "",
-        "glyphs": "https://fonts.openmaptiles.org/{fontstack}/{range}.pbf",
-        "layers": [
-            # Background
-            {
-                "id": "background",
-                "type": "background",
-                "paint": {"background-color": "#f8f4f0"},
-            },
-            # Water
-            {
-                "id": "water",
-                "type": "fill",
-                "source": "osm",
-                "source-layer": "water",
-                "paint": {"fill-color": "#a0c8f0"},
-            },
-            # Landuse - Parks
-            {
-                "id": "landuse-park",
-                "type": "fill",
-                "source": "osm",
-                "source-layer": "landuse",
-                "filter": ["==", "class", "park"],
-                "paint": {"fill-color": "#c8e6c8", "fill-opacity": 0.5},
-            },
-            # Landuse - Forest
-            {
-                "id": "landuse-forest",
-                "type": "fill",
-                "source": "osm",
-                "source-layer": "landuse",
-                "filter": ["==", "class", "wood"],
-                "paint": {"fill-color": "#94d294", "fill-opacity": 0.5},
-            },
-            # Buildings
-            {
-                "id": "building",
-                "type": "fill",
-                "source": "osm",
-                "source-layer": "building",
-                "minzoom": 13,
-                "paint": {"fill-color": "#d9d0c9", "fill-opacity": 0.8},
-            },
-            # Roads - Minor
-            {
-                "id": "road-minor",
-                "type": "line",
-                "source": "osm",
-                "source-layer": "transportation",
-                "filter": ["all", ["==", "$type", "LineString"], ["in", "class", "minor", "service"]],
-                "paint": {"line-color": "#ffffff", "line-width": 1},
-            },
-            # Roads - Major
-            {
-                "id": "road-major",
-                "type": "line",
-                "source": "osm",
-                "source-layer": "transportation",
-                "filter": ["all", ["==", "$type", "LineString"], ["in", "class", "primary", "secondary", "tertiary"]],
-                "paint": {"line-color": "#ffc107", "line-width": 2},
-            },
-            # Roads - Highway
-            {
-                "id": "road-highway",
-                "type": "line",
-                "source": "osm",
-                "source-layer": "transportation",
-                "filter": ["all", ["==", "$type", "LineString"], ["==", "class", "motorway"]],
-                "paint": {"line-color": "#ff6f00", "line-width": 3},
-            },
-            # Place labels
-            {
-                "id": "place-label",
-                "type": "symbol",
-                "source": "osm",
-                "source-layer": "place",
-                "layout": {
-                    "text-field": "{name}",
-                    "text-font": ["Open Sans Regular"],
-                    "text-size": 12,
-                },
-                "paint": {"text-color": "#333333", "text-halo-color": "#ffffff", "text-halo-width": 1},
-            },
-        ],
-        "terrain": {"source": "terrain", "exaggeration": 1.5},
-    }
-
-    return JSONResponse(content=style)
-
-
-@router.get("/metadata")
-async def get_tile_metadata():
-    """
-    Get metadata about available tiles.
-
-    Returns information about data coverage, zoom levels, and update status.
-    """
-    metadata = await tile_service.get_metadata()
-
-    return {
-        "name": "GeoEdu Germany Tiles",
-        "description": "Self-hosted OSM vector tiles for Germany",
-        "format": "pbf",
-        "scheme": "xyz",
-        "minzoom": metadata.get("minzoom", 0),
-        "maxzoom": metadata.get("maxzoom", 14),
-        "bounds": metadata.get("bounds", [5.87, 47.27, 15.04, 55.06]),  # Germany bbox
-        "center": metadata.get("center", [10.45, 51.16, 6]),  # Center of Germany
-        "attribution": "© OpenStreetMap contributors (ODbL)",
-        "data_available": metadata.get("data_available", False),
-        "last_updated": metadata.get("last_updated"),
-    }
-
-
-@router.get("/bounds")
-async def get_tile_bounds():
-    """
-    Get the geographic bounds of available tile data.
-
-    Returns bounding box for Germany in [west, south, east, north] format.
-    """
-    return {
-        "bounds": [5.87, 47.27, 15.04, 55.06],  # Germany bounding box
-        "center": [10.45, 51.16],
-        "description": "Germany (Deutschland)",
-    }
--- a/geo-service/config.py
+++ b/geo-service/config.py
@@ -1,99 +0,0 @@
-"""
-GeoEdu Service Configuration
-Environment-based configuration with Pydantic Settings
-"""
-from functools import lru_cache
-from typing import Optional
-from pydantic_settings import BaseSettings, SettingsConfigDict
-
-
-class Settings(BaseSettings):
-    """Application settings loaded from environment variables."""
-
-    model_config = SettingsConfigDict(
-        env_file=".env",
-        env_file_encoding="utf-8",
-        case_sensitive=False,
-    )
-
-    # Service Config
-    port: int = 8088
-    environment: str = "development"
-    debug: bool = False
-
-    # JWT Authentication
-    jwt_secret: str = "your-super-secret-jwt-key"
-    jwt_algorithm: str = "HS256"
-    jwt_expiration_hours: int = 24
-
-    # PostgreSQL (PostGIS)
-    database_url: str = "postgresql+asyncpg://breakpilot:breakpilot123@postgres:5432/breakpilot_db"
-
-    # MinIO (S3-compatible storage)
-    minio_endpoint: str = "minio:9000"
-    minio_access_key: str = "breakpilot"
-    minio_secret_key: str = "breakpilot123"
-    minio_bucket: str = "breakpilot-geo"
-    minio_secure: bool = False
-
-    # Ollama (LLM for learning nodes)
-    ollama_base_url: str = "http://host.docker.internal:11434"
-    ollama_model: str = "qwen2.5:14b"
-    ollama_timeout: int = 120
-
-    # Data Directories
-    osm_data_dir: str = "/app/data/osm"
-    dem_data_dir: str = "/app/data/dem"
-    tile_cache_dir: str = "/app/cache/tiles"
-    bundle_dir: str = "/app/bundles"
-
-    # Tile Server Config
-    default_pmtiles_file: str = "germany.pmtiles"
-    tile_cache_max_size_gb: float = 50.0
-
-    # DEM Config
-    dem_resolution: str = "GLO-30"  # 30m Copernicus DEM
-    terrain_tile_size: int = 256
-
-    # AOI Limits (DSGVO data minimization)
-    max_aoi_size_km2: float = 4.0  # Max 4 km² per AOI
-    max_aoi_per_user: int = 10
-    aoi_retention_days: int = 30  # Auto-delete after 30 days
-
-    # Learning Nodes
-    max_nodes_per_aoi: int = 20
-    supported_themes: list[str] = [
-        "topographie",
-        "landnutzung",
-        "orientierung",
-        "geologie",
-        "hydrologie",
-        "vegetation",
-    ]
-
-    # CORS (for frontend access)
-    cors_origins: list[str] = [
-        "http://localhost:3000",
-        "http://localhost:3001",
-        "http://localhost:8088",
-    ]
-
-    @property
-    def pmtiles_path(self) -> str:
-        """Full path to PMTiles file."""
-        return f"{self.osm_data_dir}/{self.default_pmtiles_file}"
-
-    @property
-    def is_development(self) -> bool:
-        """Check if running in development mode."""
-        return self.environment == "development"
-
-
-@lru_cache
-def get_settings() -> Settings:
-    """Get cached settings instance."""
-    return Settings()
-
-
-# Export settings instance for convenience
-settings = get_settings()
--- a/geo-service/main.py
+++ b/geo-service/main.py
@@ -1,192 +0,0 @@
-"""
-GeoEdu Service - Self-Hosted OSM + Terrain Learning Platform
-DSGVO-konforme Erdkunde-Lernplattform mit selbst gehostetem OpenStreetMap
-
-Main FastAPI Application
-"""
-import structlog
-from contextlib import asynccontextmanager
-from fastapi import FastAPI, Request
-from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import JSONResponse
-import time
-
-from config import settings
-
-# Configure structured logging
-structlog.configure(
-    processors=[
-        structlog.stdlib.filter_by_level,
-        structlog.stdlib.add_logger_name,
-        structlog.stdlib.add_log_level,
-        structlog.stdlib.PositionalArgumentsFormatter(),
-        structlog.processors.TimeStamper(fmt="iso"),
-        structlog.processors.StackInfoRenderer(),
-        structlog.processors.format_exc_info,
-        structlog.processors.UnicodeDecoder(),
-        structlog.processors.JSONRenderer() if not settings.is_development else structlog.dev.ConsoleRenderer(),
-    ],
-    wrapper_class=structlog.stdlib.BoundLogger,
-    context_class=dict,
-    logger_factory=structlog.stdlib.LoggerFactory(),
-    cache_logger_on_first_use=True,
-)
-
-logger = structlog.get_logger(__name__)
-
-
-@asynccontextmanager
-async def lifespan(app: FastAPI):
-    """Application lifespan manager."""
-    # Startup
-    logger.info(
-        "Starting GeoEdu Service",
-        environment=settings.environment,
-        port=settings.port,
-    )
-
-    # Check data directories
-    import os
-    for dir_name, dir_path in [
-        ("OSM Data", settings.osm_data_dir),
-        ("DEM Data", settings.dem_data_dir),
-        ("Tile Cache", settings.tile_cache_dir),
-        ("Bundles", settings.bundle_dir),
-    ]:
-        if os.path.exists(dir_path):
-            logger.info(f"{dir_name} directory exists", path=dir_path)
-        else:
-            logger.warning(f"{dir_name} directory missing", path=dir_path)
-            os.makedirs(dir_path, exist_ok=True)
-
-    yield
-
-    # Shutdown
-    logger.info("Shutting down GeoEdu Service")
-
-
-# Create FastAPI app
-app = FastAPI(
-    title="GeoEdu Service",
-    description="DSGVO-konforme Erdkunde-Lernplattform mit selbst gehostetem OpenStreetMap",
-    version="1.0.0",
-    docs_url="/docs" if settings.is_development else None,
-    redoc_url="/redoc" if settings.is_development else None,
-    lifespan=lifespan,
-)
-
-# CORS middleware
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=settings.cors_origins,
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"],
-)
-
-
-# Request timing middleware
-@app.middleware("http")
-async def add_timing_header(request: Request, call_next):
-    """Add X-Process-Time header to all responses."""
-    start_time = time.time()
-    response = await call_next(request)
-    process_time = time.time() - start_time
-    response.headers["X-Process-Time"] = str(process_time)
-    return response
-
-
-# Import and register routers
-from api.tiles import router as tiles_router
-from api.terrain import router as terrain_router
-from api.aoi import router as aoi_router
-from api.learning import router as learning_router
-
-app.include_router(tiles_router, prefix="/api/v1/tiles", tags=["Tiles"])
-app.include_router(terrain_router, prefix="/api/v1/terrain", tags=["Terrain"])
-app.include_router(aoi_router, prefix="/api/v1/aoi", tags=["AOI"])
-app.include_router(learning_router, prefix="/api/v1/learning", tags=["Learning"])
-
-
-# Health check endpoint
-@app.get("/health", tags=["System"])
-async def health_check():
-    """
-    Health check endpoint for Docker/Kubernetes probes.
-    Returns service status and basic metrics.
-    """
-    import os
-
-    # Check data availability
-    pmtiles_exists = os.path.exists(settings.pmtiles_path)
-    dem_exists = os.path.exists(settings.dem_data_dir) and len(os.listdir(settings.dem_data_dir)) > 0
-
-    return {
-        "status": "healthy",
-        "service": "geo-service",
-        "version": "1.0.0",
-        "environment": settings.environment,
-        "data_status": {
-            "pmtiles_available": pmtiles_exists,
-            "dem_available": dem_exists,
-            "tile_cache_dir": os.path.exists(settings.tile_cache_dir),
-            "bundle_dir": os.path.exists(settings.bundle_dir),
-        },
-        "config": {
-            "max_aoi_size_km2": settings.max_aoi_size_km2,
-            "supported_themes": settings.supported_themes,
-        },
-    }
-
-
-# Root endpoint
-@app.get("/", tags=["System"])
-async def root():
-    """Root endpoint with service information."""
-    return {
-        "service": "GeoEdu Service",
-        "description": "DSGVO-konforme Erdkunde-Lernplattform",
-        "version": "1.0.0",
-        "docs": "/docs" if settings.is_development else "disabled",
-        "endpoints": {
-            "tiles": "/api/v1/tiles",
-            "terrain": "/api/v1/terrain",
-            "aoi": "/api/v1/aoi",
-            "learning": "/api/v1/learning",
-        },
-        "attribution": {
-            "osm": "© OpenStreetMap contributors (ODbL)",
-            "dem": "© Copernicus Service (free, attribution required)",
-        },
-    }
-
-
-# Error handlers
-@app.exception_handler(404)
-async def not_found_handler(request: Request, exc):
-    """Handle 404 errors."""
-    return JSONResponse(
-        status_code=404,
-        content={"error": "Not found", "path": str(request.url.path)},
-    )
-
-
-@app.exception_handler(500)
-async def internal_error_handler(request: Request, exc):
-    """Handle 500 errors."""
-    logger.error("Internal server error", path=str(request.url.path), error=str(exc))
-    return JSONResponse(
-        status_code=500,
-        content={"error": "Internal server error"},
-    )
-
-
-if __name__ == "__main__":
-    import uvicorn
-
-    uvicorn.run(
-        "main:app",
-        host="0.0.0.0",
-        port=settings.port,
-        reload=settings.is_development,
-    )
--- a/geo-service/models/init.py
+++ b/geo-service/models/init.py
@@ -1,19 +0,0 @@
-"""
-GeoEdu Service - Pydantic Models
-"""
-from .aoi import AOIRequest, AOIResponse, AOIStatus, AOIManifest
-from .learning_node import LearningNode, LearningNodeRequest, LearningTheme, NodeType
-from .attribution import Attribution, AttributionSource
-
-__all__ = [
-    "AOIRequest",
-    "AOIResponse",
-    "AOIStatus",
-    "AOIManifest",
-    "LearningNode",
-    "LearningNodeRequest",
-    "LearningTheme",
-    "NodeType",
-    "Attribution",
-    "AttributionSource",
-]
--- a/geo-service/models/aoi.py
+++ b/geo-service/models/aoi.py
@@ -1,162 +0,0 @@
-"""
-AOI (Area of Interest) Models
-Pydantic models for AOI requests and responses
-"""
-from enum import Enum
-from typing import Optional
-from datetime import datetime
-from pydantic import BaseModel, Field
-
-
-class AOIStatus(str, Enum):
-    """Status of an AOI processing job."""
-    QUEUED = "queued"
-    PROCESSING = "processing"
-    COMPLETED = "completed"
-    FAILED = "failed"
-
-
-class AOIQuality(str, Enum):
-    """Quality level for AOI bundle generation."""
-    LOW = "low"
-    MEDIUM = "medium"
-    HIGH = "high"
-
-
-class AOITheme(str, Enum):
-    """Learning theme for AOI."""
-    TOPOGRAPHIE = "topographie"
-    LANDNUTZUNG = "landnutzung"
-    ORIENTIERUNG = "orientierung"
-    GEOLOGIE = "geologie"
-    HYDROLOGIE = "hydrologie"
-    VEGETATION = "vegetation"
-
-
-class GeoJSONPolygon(BaseModel):
-    """GeoJSON Polygon geometry."""
-    type: str = Field("Polygon", const=True)
-    coordinates: list[list[list[float]]] = Field(
-        ...,
-        description="Polygon coordinates as [[[lon, lat], ...]]",
-    )
-
-
-class AOIRequest(BaseModel):
-    """Request model for creating an AOI."""
-    polygon: dict = Field(
-        ...,
-        description="GeoJSON Polygon geometry",
-        example={
-            "type": "Polygon",
-            "coordinates": [[[9.19, 47.71], [9.20, 47.71], [9.20, 47.70], [9.19, 47.70], [9.19, 47.71]]]
-        },
-    )
-    theme: str = Field(
-        "topographie",
-        description="Learning theme for the AOI",
-    )
-    quality: str = Field(
-        "medium",
-        pattern="^(low|medium|high)$",
-        description="Bundle quality level",
-    )
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "polygon": {
-                    "type": "Polygon",
-                    "coordinates": [[[9.1875, 47.7055], [9.1975, 47.7055], [9.1975, 47.7115], [9.1875, 47.7115], [9.1875, 47.7055]]]
-                },
-                "theme": "topographie",
-                "quality": "medium",
-            }
-        }
-
-
-class AOIResponse(BaseModel):
-    """Response model for AOI operations."""
-    aoi_id: str = Field(..., description="Unique AOI identifier")
-    status: AOIStatus = Field(..., description="Current processing status")
-    area_km2: float = Field(0, description="Area in square kilometers")
-    estimated_size_mb: float = Field(0, description="Estimated bundle size in MB")
-    message: Optional[str] = Field(None, description="Status message")
-    download_url: Optional[str] = Field(None, description="Bundle download URL")
-    manifest_url: Optional[str] = Field(None, description="Manifest URL")
-    created_at: Optional[str] = Field(None, description="Creation timestamp")
-    completed_at: Optional[str] = Field(None, description="Completion timestamp")
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "aoi_id": "550e8400-e29b-41d4-a716-446655440000",
-                "status": "completed",
-                "area_km2": 0.45,
-                "estimated_size_mb": 25.5,
-                "message": "AOI processing complete",
-                "download_url": "/api/v1/aoi/550e8400-e29b-41d4-a716-446655440000/bundle.zip",
-                "manifest_url": "/api/v1/aoi/550e8400-e29b-41d4-a716-446655440000/manifest.json",
-            }
-        }
-
-
-class AOIBounds(BaseModel):
-    """Geographic bounding box."""
-    west: float = Field(..., description="Western longitude")
-    south: float = Field(..., description="Southern latitude")
-    east: float = Field(..., description="Eastern longitude")
-    north: float = Field(..., description="Northern latitude")
-
-
-class AOICenter(BaseModel):
-    """Geographic center point."""
-    longitude: float
-    latitude: float
-
-
-class AOIManifest(BaseModel):
-    """Unity bundle manifest for an AOI."""
-    version: str = Field("1.0.0", description="Manifest version")
-    aoi_id: str = Field(..., description="AOI identifier")
-    created_at: str = Field(..., description="Creation timestamp")
-    bounds: AOIBounds = Field(..., description="Geographic bounds")
-    center: AOICenter = Field(..., description="Geographic center")
-    area_km2: float = Field(..., description="Area in km²")
-    theme: str = Field(..., description="Learning theme")
-    quality: str = Field(..., description="Quality level")
-    assets: dict = Field(..., description="Asset file references")
-    unity: dict = Field(..., description="Unity-specific configuration")
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "version": "1.0.0",
-                "aoi_id": "550e8400-e29b-41d4-a716-446655440000",
-                "created_at": "2024-01-15T12:00:00Z",
-                "bounds": {
-                    "west": 9.1875,
-                    "south": 47.7055,
-                    "east": 9.1975,
-                    "north": 47.7115,
-                },
-                "center": {
-                    "longitude": 9.1925,
-                    "latitude": 47.7085,
-                },
-                "area_km2": 0.45,
-                "theme": "topographie",
-                "quality": "medium",
-                "assets": {
-                    "terrain": {"file": "terrain.heightmap.png", "config": "terrain.json"},
-                    "osm_features": {"file": "osm_features.json"},
-                    "learning_positions": {"file": "learning_positions.json"},
-                    "attribution": {"file": "attribution.json"},
-                },
-                "unity": {
-                    "coordinate_system": "Unity (Y-up, left-handed)",
-                    "scale": 1.0,
-                    "terrain_resolution": 256,
-                },
-            }
-        }
--- a/geo-service/models/attribution.py
+++ b/geo-service/models/attribution.py
@@ -1,97 +0,0 @@
-"""
-Attribution Models
-Models for license and attribution tracking
-"""
-from typing import Optional
-from pydantic import BaseModel, Field
-
-
-class AttributionSource(BaseModel):
-    """
-    Attribution information for a data source.
-
-    All geographic data requires proper attribution per their licenses.
-    """
-    name: str = Field(..., description="Source name")
-    license: str = Field(..., description="License name")
-    url: str = Field(..., description="License or source URL")
-    attribution: str = Field(..., description="Required attribution text")
-    required: bool = Field(True, description="Whether attribution is legally required")
-    logo_url: Optional[str] = Field(None, description="Optional logo URL")
-
-
-class Attribution(BaseModel):
-    """
-    Complete attribution information for an AOI bundle.
-
-    Ensures DSGVO/GDPR compliance and proper data source attribution.
-    """
-    sources: list[AttributionSource] = Field(
-        ...,
-        description="List of data sources requiring attribution",
-    )
-    generated_at: str = Field(..., description="Timestamp when attribution was generated")
-    notice: str = Field(
-        "This data must be attributed according to the licenses above when used publicly.",
-        description="General attribution notice",
-    )
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "sources": [
-                    {
-                        "name": "OpenStreetMap",
-                        "license": "Open Database License (ODbL) v1.0",
-                        "url": "https://www.openstreetmap.org/copyright",
-                        "attribution": "© OpenStreetMap contributors",
-                        "required": True,
-                    },
-                    {
-                        "name": "Copernicus DEM",
-                        "license": "Copernicus Data License",
-                        "url": "https://spacedata.copernicus.eu/",
-                        "attribution": "© Copernicus Service Information 2024",
-                        "required": True,
-                    },
-                ],
-                "generated_at": "2024-01-15T12:00:00Z",
-                "notice": "This data must be attributed according to the licenses above when used publicly.",
-            }
-        }
-
-
-# Predefined attribution sources
-OSM_ATTRIBUTION = AttributionSource(
-    name="OpenStreetMap",
-    license="Open Database License (ODbL) v1.0",
-    url="https://www.openstreetmap.org/copyright",
-    attribution="© OpenStreetMap contributors",
-    required=True,
-)
-
-COPERNICUS_ATTRIBUTION = AttributionSource(
-    name="Copernicus DEM",
-    license="Copernicus Data License",
-    url="https://spacedata.copernicus.eu/",
-    attribution="© Copernicus Service Information 2024",
-    required=True,
-)
-
-OPENAERIAL_ATTRIBUTION = AttributionSource(
-    name="OpenAerialMap",
-    license="CC-BY 4.0",
-    url="https://openaerialmap.org/",
-    attribution="© OpenAerialMap contributors",
-    required=True,
-)
-
-
-def get_default_attribution() -> Attribution:
-    """Get default attribution with standard sources."""
-    from datetime import datetime
-
-    return Attribution(
-        sources=[OSM_ATTRIBUTION, COPERNICUS_ATTRIBUTION],
-        generated_at=datetime.utcnow().isoformat(),
-    )
--- a/geo-service/models/learning_node.py
+++ b/geo-service/models/learning_node.py
@@ -1,120 +0,0 @@
-"""
-Learning Node Models
-Pydantic models for educational content nodes
-"""
-from enum import Enum
-from typing import Optional
-from pydantic import BaseModel, Field
-
-
-class LearningTheme(str, Enum):
-    """Available learning themes for geographic education."""
-    TOPOGRAPHIE = "topographie"
-    LANDNUTZUNG = "landnutzung"
-    ORIENTIERUNG = "orientierung"
-    GEOLOGIE = "geologie"
-    HYDROLOGIE = "hydrologie"
-    VEGETATION = "vegetation"
-
-
-class NodeType(str, Enum):
-    """Type of learning node interaction."""
-    QUESTION = "question"  # Multiple choice or open question
-    OBSERVATION = "observation"  # Guided observation task
-    EXPLORATION = "exploration"  # Free exploration with hints
-
-
-class Position(BaseModel):
-    """Geographic position for a learning node."""
-    latitude: float = Field(..., ge=-90, le=90, description="Latitude in degrees")
-    longitude: float = Field(..., ge=-180, le=180, description="Longitude in degrees")
-    altitude: Optional[float] = Field(None, description="Altitude in meters")
-
-
-class LearningNode(BaseModel):
-    """
-    A learning node (station) within a geographic area.
-
-    Contains educational content tied to a specific location,
-    including questions, hints, and explanations.
-    """
-    id: str = Field(..., description="Unique node identifier")
-    aoi_id: str = Field(..., description="Parent AOI identifier")
-    title: str = Field(..., min_length=1, max_length=100, description="Node title")
-    theme: LearningTheme = Field(..., description="Learning theme")
-    position: dict = Field(..., description="Geographic position")
-    question: str = Field(..., description="Learning question or task")
-    hints: list[str] = Field(default_factory=list, description="Progressive hints")
-    answer: str = Field(..., description="Correct answer or expected observation")
-    explanation: str = Field(..., description="Didactic explanation")
-    node_type: NodeType = Field(NodeType.QUESTION, description="Interaction type")
-    points: int = Field(10, ge=1, le=100, description="Points awarded for completion")
-    approved: bool = Field(False, description="Teacher-approved for student use")
-    media: Optional[list[dict]] = Field(None, description="Associated media files")
-    tags: Optional[list[str]] = Field(None, description="Content tags")
-    difficulty: Optional[str] = Field(None, description="Difficulty level")
-    grade_level: Optional[str] = Field(None, description="Target grade level")
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "id": "node-001",
-                "aoi_id": "550e8400-e29b-41d4-a716-446655440000",
-                "title": "Höhenbestimmung",
-                "theme": "topographie",
-                "position": {"latitude": 47.7085, "longitude": 9.1925},
-                "question": "Schätze die Höhe dieses Punktes über dem Meeresspiegel.",
-                "hints": [
-                    "Schau dir die Vegetation an.",
-                    "Vergleiche mit dem Seespiegel des Bodensees (395m).",
-                ],
-                "answer": "Ca. 430 Meter über NN",
-                "explanation": "Die Höhe lässt sich aus der Vegetation und der relativen Position zum Bodensee abschätzen.",
-                "node_type": "question",
-                "points": 10,
-                "approved": True,
-            }
-        }
-
-
-class LearningNodeRequest(BaseModel):
-    """Request model for creating a learning node manually."""
-    title: str = Field(..., min_length=1, max_length=100)
-    theme: LearningTheme
-    position: Position
-    question: str
-    hints: list[str] = Field(default_factory=list)
-    answer: str
-    explanation: str
-    node_type: NodeType = NodeType.QUESTION
-    points: int = Field(10, ge=1, le=100)
-    tags: Optional[list[str]] = None
-    difficulty: Optional[str] = Field(None, pattern="^(leicht|mittel|schwer)$")
-    grade_level: Optional[str] = None
-
-
-class LearningNodeBatch(BaseModel):
-    """Batch of learning nodes for bulk operations."""
-    nodes: list[LearningNode]
-    total_points: int = 0
-
-    def calculate_total_points(self) -> int:
-        """Calculate total points from all nodes."""
-        self.total_points = sum(node.points for node in self.nodes)
-        return self.total_points
-
-
-class LearningProgress(BaseModel):
-    """Student progress through learning nodes."""
-    student_id: str
-    aoi_id: str
-    completed_nodes: list[str] = Field(default_factory=list)
-    total_points: int = 0
-    started_at: Optional[str] = None
-    completed_at: Optional[str] = None
-
-    @property
-    def completion_percentage(self) -> float:
-        """Calculate completion percentage."""
-        # Would need total node count for accurate calculation
-        return len(self.completed_nodes) * 10  # Placeholder
--- a/geo-service/requirements.txt
+++ b/geo-service/requirements.txt
@@ -1,45 +0,0 @@
-# FastAPI Framework
-fastapi==0.115.0
-uvicorn[standard]==0.30.6
-python-multipart==0.0.9
-
-# Database & GIS
-asyncpg==0.29.0
-sqlalchemy[asyncio]==2.0.32
-geoalchemy2==0.14.7
-shapely==2.0.5
-pyproj==3.6.1
-
-# Geo Processing
-rasterio==1.3.10
-numpy==1.26.4
-pillow==10.4.0
-
-# PMTiles Support
-pmtiles==3.2.0
-
-# MinIO/S3 Client
-minio==7.2.7
-boto3==1.34.149
-
-# HTTP Client (for Ollama)
-httpx==0.27.0
-aiohttp==3.10.4
-
-# Validation & Settings
-pydantic==2.8.2
-pydantic-settings==2.4.0
-python-dotenv==1.0.1
-
-# Authentication
-python-jose[cryptography]==3.3.0
-passlib[bcrypt]==1.7.4
-
-# Utilities
-orjson==3.10.6
-structlog==24.4.0
-
-# Testing
-pytest==8.3.2
-pytest-asyncio==0.23.8
-httpx==0.27.0
--- a/geo-service/scripts/download_dem.sh
+++ b/geo-service/scripts/download_dem.sh
@@ -1,198 +0,0 @@
-#!/bin/bash
-# ============================================
-# Copernicus DEM Download Script for GeoEdu Service
-# ============================================
-#
-# WICHTIG: Dieses Script startet einen Download von ca. 20-40 GB!
-# Nur nach expliziter Freigabe ausfuehren!
-#
-# Quelle: Copernicus Data Space Ecosystem
-# Daten: GLO-30 DEM (30m Aufloesung)
-# Lizenz: Copernicus Data (frei, Attribution erforderlich)
-# Attribution: © Copernicus Service Information
-#
-# Voraussetzungen:
-# - Copernicus Data Space Account (kostenlos)
-# - Credentials in ~/.netrc oder als Umgebungsvariablen
-#
-# Nutzung:
-#   ./download_dem.sh [--dry-run] [--bbox west,south,east,north]
-#
-# Beispiel (nur Bayern):
-#   ./download_dem.sh --bbox 8.97,47.27,13.84,50.56
-#
-# ============================================
-
-set -e
-
-# Configuration
-DATA_DIR="${DEM_DATA_DIR:-/app/data/dem}"
-COPERNICUS_URL="https://dataspace.copernicus.eu"
-
-# Germany bounding box (default)
-BBOX_WEST="${BBOX_WEST:-5.87}"
-BBOX_SOUTH="${BBOX_SOUTH:-47.27}"
-BBOX_EAST="${BBOX_EAST:-15.04}"
-BBOX_NORTH="${BBOX_NORTH:-55.06}"
-
-# Colors for output
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-NC='\033[0m' # No Color
-
-echo -e "${YELLOW}============================================${NC}"
-echo -e "${YELLOW}GeoEdu Service - Copernicus DEM Download${NC}"
-echo -e "${YELLOW}============================================${NC}"
-echo ""
-echo -e "Quelle:      ${GREEN}Copernicus Data Space${NC}"
-echo -e "Daten:       ${GREEN}GLO-30 DEM (30m Aufloesung)${NC}"
-echo -e "Groesse:     ${YELLOW}~20-40 GB (Deutschland komplett)${NC}"
-echo -e "Lizenz:      ${GREEN}Copernicus Data (frei)${NC}"
-echo -e "Attribution: ${GREEN}© Copernicus Service Information${NC}"
-echo ""
-
-# Parse arguments
-DRY_RUN=false
-while [[ $# -gt 0 ]]; do
-    case $1 in
-        --dry-run)
-            DRY_RUN=true
-            shift
-            ;;
-        --bbox)
-            IFS=',' read -r BBOX_WEST BBOX_SOUTH BBOX_EAST BBOX_NORTH <<< "$2"
-            shift 2
-            ;;
-        *)
-            echo "Unbekannte Option: $1"
-            exit 1
-            ;;
-    esac
-done
-
-echo "Bounding Box:"
-echo "  West:  $BBOX_WEST"
-echo "  Sued:  $BBOX_SOUTH"
-echo "  Ost:   $BBOX_EAST"
-echo "  Nord:  $BBOX_NORTH"
-echo ""
-
-# Calculate required tiles
-# Copernicus DEM tiles are 1°x1° cells
-calc_tiles() {
-    local west=$(echo "$BBOX_WEST" | cut -d. -f1)
-    local south=$(echo "$BBOX_SOUTH" | cut -d. -f1)
-    local east=$(echo "$BBOX_EAST" | cut -d. -f1)
-    local north=$(echo "$BBOX_NORTH" | cut -d. -f1)
-
-    # Adjust for negative values
-    west=$((west < 0 ? west : west))
-
-    local count=0
-    for lat in $(seq $south $north); do
-        for lon in $(seq $west $east); do
-            count=$((count + 1))
-        done
-    done
-    echo $count
-}
-
-TILE_COUNT=$(calc_tiles)
-ESTIMATED_SIZE=$((TILE_COUNT * 50)) # ~50 MB per tile
-
-echo "Benoetigte Tiles: $TILE_COUNT"
-echo "Geschaetzte Groesse: ~${ESTIMATED_SIZE} MB"
-echo ""
-
-if [ "$DRY_RUN" = true ]; then
-    echo -e "${YELLOW}[DRY-RUN] Kein Download wird durchgefuehrt.${NC}"
-    echo ""
-    echo "Wuerde folgende Tiles herunterladen:"
-    echo ""
-
-    for lat in $(seq $(echo "$BBOX_SOUTH" | cut -d. -f1) $(echo "$BBOX_NORTH" | cut -d. -f1)); do
-        for lon in $(seq $(echo "$BBOX_WEST" | cut -d. -f1) $(echo "$BBOX_EAST" | cut -d. -f1)); do
-            lat_prefix=$([ $lat -ge 0 ] && echo "N" || echo "S")
-            lon_prefix=$([ $lon -ge 0 ] && echo "E" || echo "W")
-            printf "  %s%02d%s%03d.tif\n" "$lat_prefix" "${lat#-}" "$lon_prefix" "${lon#-}"
-        done
-    done
-    echo ""
-    exit 0
-fi
-
-# Confirm download
-echo -e "${RED}ACHTUNG: Download startet ~${ESTIMATED_SIZE} MB Daten!${NC}"
-echo ""
-read -p "Download starten? (j/N) " -n 1 -r
-echo ""
-
-if [[ ! $REPLY =~ ^[JjYy]$ ]]; then
-    echo "Download abgebrochen."
-    exit 1
-fi
-
-# Create data directory
-mkdir -p "$DATA_DIR"
-cd "$DATA_DIR"
-
-# Check for AWS CLI (preferred method for Copernicus)
-if ! command -v aws &> /dev/null; then
-    echo -e "${YELLOW}HINWEIS: AWS CLI nicht installiert.${NC}"
-    echo "Installiere mit: pip install awscli"
-    echo ""
-    echo "Alternative: Manueller Download von:"
-    echo "  https://dataspace.copernicus.eu/explore-data/data-collections/copernicus-contributing-missions/collections-description/COP-DEM"
-    exit 1
-fi
-
-# Download tiles
-echo ""
-echo -e "${GREEN}Starte Download...${NC}"
-
-DOWNLOADED=0
-FAILED=0
-
-for lat in $(seq $(echo "$BBOX_SOUTH" | cut -d. -f1) $(echo "$BBOX_NORTH" | cut -d. -f1)); do
-    for lon in $(seq $(echo "$BBOX_WEST" | cut -d. -f1) $(echo "$BBOX_EAST" | cut -d. -f1)); do
-        lat_prefix=$([ $lat -ge 0 ] && echo "N" || echo "S")
-        lon_prefix=$([ $lon -ge 0 ] && echo "E" || echo "W")
-
-        lat_abs=${lat#-}
-        lon_abs=${lon#-}
-
-        filename=$(printf "%s%02d%s%03d.tif" "$lat_prefix" "$lat_abs" "$lon_prefix" "$lon_abs")
-
-        if [ -f "$filename" ]; then
-            echo "  ✓ $filename (bereits vorhanden)"
-            DOWNLOADED=$((DOWNLOADED + 1))
-            continue
-        fi
-
-        echo "  ↓ $filename"
-
-        # Copernicus DEM S3 bucket path
-        # Format: s3://copernicus-dem-30m/Copernicus_DSM_COG_10_N47_00_E008_00_DEM/
-        s3_path="s3://copernicus-dem-30m/Copernicus_DSM_COG_10_${lat_prefix}${lat_abs}_00_${lon_prefix}${lon_abs}_00_DEM/"
-
-        if aws s3 cp "${s3_path}Copernicus_DSM_COG_10_${lat_prefix}${lat_abs}_00_${lon_prefix}${lon_abs}_00_DEM.tif" "$filename" --no-sign-request 2>/dev/null; then
-            DOWNLOADED=$((DOWNLOADED + 1))
-        else
-            echo -e "    ${YELLOW}(nicht verfuegbar)${NC}"
-            FAILED=$((FAILED + 1))
-        fi
-    done
-done
-
-echo ""
-echo -e "${GREEN}============================================${NC}"
-echo -e "${GREEN}Download abgeschlossen${NC}"
-echo -e "${GREEN}============================================${NC}"
-echo ""
-echo "Heruntergeladen: $DOWNLOADED Tiles"
-echo "Nicht verfuegbar: $FAILED Tiles"
-echo "Speicherort: $DATA_DIR"
-echo ""
-echo -e "${YELLOW}Naechster Schritt:${NC}"
-echo "  Die Tiles werden automatisch vom geo-service geladen."
--- a/geo-service/scripts/download_osm.sh
+++ b/geo-service/scripts/download_osm.sh
@@ -1,113 +0,0 @@
-#!/bin/bash
-# ============================================
-# OSM Data Download Script for GeoEdu Service
-# ============================================
-#
-# WICHTIG: Dieses Script startet einen Download von ca. 4.4 GB!
-# Nur nach expliziter Freigabe ausfuehren!
-#
-# Quelle: Geofabrik (offizieller OSM-Mirror)
-# Lizenz: ODbL (Open Database License)
-# Attribution: © OpenStreetMap contributors
-#
-# Nutzung:
-#   ./download_osm.sh [--dry-run]
-#
-# ============================================
-
-set -e
-
-# Configuration
-DATA_DIR="${OSM_DATA_DIR:-/app/data/osm}"
-GEOFABRIK_URL="https://download.geofabrik.de/europe/germany-latest.osm.pbf"
-GEOFABRIK_MD5_URL="https://download.geofabrik.de/europe/germany-latest.osm.pbf.md5"
-OUTPUT_FILE="germany-latest.osm.pbf"
-
-# Colors for output
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-NC='\033[0m' # No Color
-
-echo -e "${YELLOW}============================================${NC}"
-echo -e "${YELLOW}GeoEdu Service - OSM Data Download${NC}"
-echo -e "${YELLOW}============================================${NC}"
-echo ""
-echo -e "Quelle:      ${GREEN}Geofabrik${NC}"
-echo -e "Datei:       ${GREEN}Germany PBF${NC}"
-echo -e "Groesse:     ${YELLOW}~4.4 GB${NC}"
-echo -e "Lizenz:      ${GREEN}ODbL (Open Database License)${NC}"
-echo -e "Attribution: ${GREEN}© OpenStreetMap contributors${NC}"
-echo ""
-
-# Check for dry-run mode
-if [[ "$1" == "--dry-run" ]]; then
-    echo -e "${YELLOW}[DRY-RUN] Kein Download wird durchgefuehrt.${NC}"
-    echo ""
-    echo "Wuerde herunterladen von:"
-    echo "  $GEOFABRIK_URL"
-    echo ""
-    echo "Nach:"
-    echo "  $DATA_DIR/$OUTPUT_FILE"
-    echo ""
-    exit 0
-fi
-
-# Confirm download
-echo -e "${RED}ACHTUNG: Download startet ~4.4 GB Daten!${NC}"
-echo ""
-read -p "Download starten? (j/N) " -n 1 -r
-echo ""
-
-if [[ ! $REPLY =~ ^[JjYy]$ ]]; then
-    echo "Download abgebrochen."
-    exit 1
-fi
-
-# Create data directory
-mkdir -p "$DATA_DIR"
-cd "$DATA_DIR"
-
-# Download MD5 checksum first
-echo ""
-echo -e "${GREEN}[1/3] Lade MD5-Pruefsumme herunter...${NC}"
-curl -L -o "${OUTPUT_FILE}.md5" "$GEOFABRIK_MD5_URL"
-
-# Download OSM data
-echo ""
-echo -e "${GREEN}[2/3] Lade OSM-Daten herunter (~4.4 GB)...${NC}"
-echo "       Dies kann je nach Internetverbindung 5-60 Minuten dauern."
-echo ""
-
-# Use wget with resume support
-if command -v wget &> /dev/null; then
-    wget -c -O "$OUTPUT_FILE" "$GEOFABRIK_URL"
-else
-    curl -L -C - -o "$OUTPUT_FILE" "$GEOFABRIK_URL"
-fi
-
-# Verify checksum
-echo ""
-echo -e "${GREEN}[3/3] Verifiziere MD5-Pruefsumme...${NC}"
-
-if md5sum -c "${OUTPUT_FILE}.md5"; then
-    echo ""
-    echo -e "${GREEN}✓ Download erfolgreich!${NC}"
-    echo ""
-    echo "Datei: $DATA_DIR/$OUTPUT_FILE"
-    echo "Groesse: $(du -h "$OUTPUT_FILE" | cut -f1)"
-    echo ""
-    echo -e "${YELLOW}Naechster Schritt:${NC}"
-    echo "  ./import_osm.sh  # OSM in PostGIS importieren"
-    echo "  ./generate_tiles.sh  # PMTiles generieren"
-else
-    echo ""
-    echo -e "${RED}✗ MD5-Pruefsumme stimmt nicht ueberein!${NC}"
-    echo "Bitte Download erneut starten."
-    exit 1
-fi
-
-echo ""
-echo -e "${YELLOW}============================================${NC}"
-echo -e "${YELLOW}Download abgeschlossen${NC}"
-echo -e "${YELLOW}============================================${NC}"
--- a/geo-service/scripts/generate_tiles.sh
+++ b/geo-service/scripts/generate_tiles.sh
@@ -1,184 +0,0 @@
-#!/bin/bash
-# ============================================
-# PMTiles Generation Script for GeoEdu Service
-# ============================================
-#
-# Generiert PMTiles Vector Tiles aus PostGIS OSM Daten
-# Verwendet tippecanoe fuer die Tile-Generierung
-#
-# Voraussetzungen:
-# - OSM Daten in PostGIS importiert
-# - tippecanoe installiert
-# - ogr2ogr (GDAL) installiert
-#
-# Nutzung:
-#   ./generate_tiles.sh [--min-zoom 0] [--max-zoom 14]
-#
-# ============================================
-
-set -e
-
-# Configuration
-DATA_DIR="${OSM_DATA_DIR:-/app/data/osm}"
-OUTPUT_FILE="${DATA_DIR}/germany.pmtiles"
-DATABASE_URL="${DATABASE_URL:-postgresql://breakpilot:breakpilot123@postgres:5432/breakpilot_db}"
-
-MIN_ZOOM="${MIN_ZOOM:-0}"
-MAX_ZOOM="${MAX_ZOOM:-14}"
-
-# Parse DATABASE_URL
-DB_USER=$(echo $DATABASE_URL | sed -n 's|.*://\([^:]*\):.*|\1|p')
-DB_PASS=$(echo $DATABASE_URL | sed -n 's|.*://[^:]*:\([^@]*\)@.*|\1|p')
-DB_HOST=$(echo $DATABASE_URL | sed -n 's|.*@\([^:]*\):.*|\1|p')
-DB_PORT=$(echo $DATABASE_URL | sed -n 's|.*:\([0-9]*\)/.*|\1|p')
-DB_NAME=$(echo $DATABASE_URL | sed -n 's|.*/\([^?]*\).*|\1|p')
-
-# Colors
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-NC='\033[0m'
-
-echo -e "${YELLOW}============================================${NC}"
-echo -e "${YELLOW}GeoEdu Service - PMTiles Generation${NC}"
-echo -e "${YELLOW}============================================${NC}"
-echo ""
-
-# Parse arguments
-while [[ $# -gt 0 ]]; do
-    case $1 in
-        --min-zoom)
-            MIN_ZOOM="$2"
-            shift 2
-            ;;
-        --max-zoom)
-            MAX_ZOOM="$2"
-            shift 2
-            ;;
-        *)
-            echo "Unbekannte Option: $1"
-            exit 1
-            ;;
-    esac
-done
-
-echo "Zoom Level: $MIN_ZOOM - $MAX_ZOOM"
-echo "Output: $OUTPUT_FILE"
-echo ""
-
-# Estimate output size
-if [ "$MAX_ZOOM" -le 14 ]; then
-    ESTIMATED_SIZE="200-500 GB"
-elif [ "$MAX_ZOOM" -le 16 ]; then
-    ESTIMATED_SIZE="500-800 GB"
-else
-    ESTIMATED_SIZE="2-4 TB"
-fi
-
-echo -e "${YELLOW}Geschaetzte Groesse: $ESTIMATED_SIZE${NC}"
-echo -e "${YELLOW}Geschaetzte Dauer: 12-24 Stunden (Zoom 0-14)${NC}"
-echo ""
-
-# Check for required tools
-for tool in tippecanoe ogr2ogr; do
-    if ! command -v $tool &> /dev/null; then
-        echo -e "${RED}Fehler: $tool nicht installiert!${NC}"
-        echo ""
-        if [ "$tool" == "tippecanoe" ]; then
-            echo "Installation:"
-            echo "  git clone https://github.com/felt/tippecanoe.git"
-            echo "  cd tippecanoe && make -j && sudo make install"
-        else
-            echo "Installation:"
-            echo "  apt-get install gdal-bin  # Debian/Ubuntu"
-            echo "  brew install gdal         # macOS"
-        fi
-        exit 1
-    fi
-done
-
-echo "tippecanoe: $(tippecanoe --version 2>&1 | head -1)"
-echo "ogr2ogr: $(ogr2ogr --version | head -1)"
-echo ""
-
-read -p "Tile-Generierung starten? (j/N) " -n 1 -r
-echo ""
-
-if [[ ! $REPLY =~ ^[JjYy]$ ]]; then
-    echo "Abgebrochen."
-    exit 1
-fi
-
-# Create temporary directory for GeoJSON
-TEMP_DIR=$(mktemp -d)
-trap "rm -rf $TEMP_DIR" EXIT
-
-echo ""
-echo -e "${GREEN}[1/5] Exportiere Landuse/Landcover...${NC}"
-
-PGPASSWORD="$DB_PASS" ogr2ogr -f GeoJSONSeq \
-    "$TEMP_DIR/landuse.geojsonseq" \
-    "PG:host=$DB_HOST port=$DB_PORT user=$DB_USER dbname=$DB_NAME password=$DB_PASS" \
-    -sql "SELECT way AS geometry, landuse, natural, name FROM planet_osm_polygon WHERE landuse IS NOT NULL OR natural IS NOT NULL"
-
-echo -e "${GREEN}[2/5] Exportiere Gebaude...${NC}"
-
-PGPASSWORD="$DB_PASS" ogr2ogr -f GeoJSONSeq \
-    "$TEMP_DIR/building.geojsonseq" \
-    "PG:host=$DB_HOST port=$DB_PORT user=$DB_USER dbname=$DB_NAME password=$DB_PASS" \
-    -sql "SELECT way AS geometry, building, name, addr_housenumber FROM planet_osm_polygon WHERE building IS NOT NULL"
-
-echo -e "${GREEN}[3/5] Exportiere Strassen...${NC}"
-
-PGPASSWORD="$DB_PASS" ogr2ogr -f GeoJSONSeq \
-    "$TEMP_DIR/transportation.geojsonseq" \
-    "PG:host=$DB_HOST port=$DB_PORT user=$DB_USER dbname=$DB_NAME password=$DB_PASS" \
-    -sql "SELECT way AS geometry, highway, railway, name, ref FROM planet_osm_line WHERE highway IS NOT NULL OR railway IS NOT NULL"
-
-echo -e "${GREEN}[4/5] Exportiere Gewaesser...${NC}"
-
-PGPASSWORD="$DB_PASS" ogr2ogr -f GeoJSONSeq \
-    "$TEMP_DIR/water.geojsonseq" \
-    "PG:host=$DB_HOST port=$DB_PORT user=$DB_USER dbname=$DB_NAME password=$DB_PASS" \
-    -sql "SELECT way AS geometry, waterway, water, name FROM planet_osm_polygon WHERE water IS NOT NULL OR waterway IS NOT NULL
-          UNION ALL
-          SELECT way AS geometry, waterway, NULL as water, name FROM planet_osm_line WHERE waterway IS NOT NULL"
-
-echo -e "${GREEN}[5/5] Exportiere Orte (POIs)...${NC}"
-
-PGPASSWORD="$DB_PASS" ogr2ogr -f GeoJSONSeq \
-    "$TEMP_DIR/place.geojsonseq" \
-    "PG:host=$DB_HOST port=$DB_PORT user=$DB_USER dbname=$DB_NAME password=$DB_PASS" \
-    -sql "SELECT way AS geometry, place, name, population FROM planet_osm_point WHERE place IS NOT NULL"
-
-echo ""
-echo -e "${GREEN}Generiere PMTiles...${NC}"
-echo "Dies kann mehrere Stunden dauern!"
-echo ""
-
-# Run tippecanoe
-tippecanoe \
-    --output="$OUTPUT_FILE" \
-    --force \
-    --name="GeoEdu Germany" \
-    --description="Self-hosted OSM tiles for DSGVO-compliant education" \
-    --attribution="© OpenStreetMap contributors" \
-    --minimum-zoom="$MIN_ZOOM" \
-    --maximum-zoom="$MAX_ZOOM" \
-    --drop-densest-as-needed \
-    --extend-zooms-if-still-dropping \
-    --layer=landuse:"$TEMP_DIR/landuse.geojsonseq" \
-    --layer=building:"$TEMP_DIR/building.geojsonseq" \
-    --layer=transportation:"$TEMP_DIR/transportation.geojsonseq" \
-    --layer=water:"$TEMP_DIR/water.geojsonseq" \
-    --layer=place:"$TEMP_DIR/place.geojsonseq"
-
-echo ""
-echo -e "${GREEN}============================================${NC}"
-echo -e "${GREEN}PMTiles Generierung abgeschlossen!${NC}"
-echo -e "${GREEN}============================================${NC}"
-echo ""
-echo "Ausgabe: $OUTPUT_FILE"
-echo "Groesse: $(du -h "$OUTPUT_FILE" | cut -f1)"
-echo ""
-echo "Die Tiles sind jetzt bereit fuer den geo-service."
--- a/geo-service/scripts/import_osm.sh
+++ b/geo-service/scripts/import_osm.sh
@@ -1,236 +0,0 @@
-#!/bin/bash
-# ============================================
-# OSM PostGIS Import Script for GeoEdu Service
-# ============================================
-#
-# Importiert OSM PBF-Daten in PostgreSQL/PostGIS
-# Verwendet osm2pgsql fuer den Import
-#
-# Voraussetzungen:
-# - PostgreSQL mit PostGIS Extension
-# - osm2pgsql installiert
-# - OSM PBF Datei heruntergeladen
-#
-# Nutzung:
-#   ./import_osm.sh [--slim] [--drop]
-#
-# ============================================
-
-set -e
-
-# Configuration
-DATA_DIR="${OSM_DATA_DIR:-/app/data/osm}"
-PBF_FILE="${DATA_DIR}/germany-latest.osm.pbf"
-DATABASE_URL="${DATABASE_URL:-postgresql://breakpilot:breakpilot123@postgres:5432/breakpilot_db}"
-
-# Parse DATABASE_URL
-DB_USER=$(echo $DATABASE_URL | sed -n 's|.*://\([^:]*\):.*|\1|p')
-DB_PASS=$(echo $DATABASE_URL | sed -n 's|.*://[^:]*:\([^@]*\)@.*|\1|p')
-DB_HOST=$(echo $DATABASE_URL | sed -n 's|.*@\([^:]*\):.*|\1|p')
-DB_PORT=$(echo $DATABASE_URL | sed -n 's|.*:\([0-9]*\)/.*|\1|p')
-DB_NAME=$(echo $DATABASE_URL | sed -n 's|.*/\([^?]*\).*|\1|p')
-
-# Colors
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-NC='\033[0m'
-
-echo -e "${YELLOW}============================================${NC}"
-echo -e "${YELLOW}GeoEdu Service - OSM PostGIS Import${NC}"
-echo -e "${YELLOW}============================================${NC}"
-echo ""
-
-# Check if PBF file exists
-if [ ! -f "$PBF_FILE" ]; then
-    echo -e "${RED}Fehler: OSM PBF Datei nicht gefunden!${NC}"
-    echo "Erwartet: $PBF_FILE"
-    echo ""
-    echo "Bitte zuerst download_osm.sh ausfuehren."
-    exit 1
-fi
-
-echo "PBF Datei: $PBF_FILE"
-echo "Groesse: $(du -h "$PBF_FILE" | cut -f1)"
-echo ""
-echo "Datenbank: $DB_NAME @ $DB_HOST:$DB_PORT"
-echo ""
-
-# Check for osm2pgsql
-if ! command -v osm2pgsql &> /dev/null; then
-    echo -e "${RED}Fehler: osm2pgsql nicht installiert!${NC}"
-    echo ""
-    echo "Installation:"
-    echo "  apt-get install osm2pgsql  # Debian/Ubuntu"
-    echo "  brew install osm2pgsql     # macOS"
-    exit 1
-fi
-
-OSM2PGSQL_VERSION=$(osm2pgsql --version 2>&1 | head -1)
-echo "osm2pgsql Version: $OSM2PGSQL_VERSION"
-echo ""
-
-# Parse arguments
-SLIM_MODE=""
-DROP_MODE=""
-while [[ $# -gt 0 ]]; do
-    case $1 in
-        --slim)
-            SLIM_MODE="--slim"
-            echo "Modus: Slim (fuer Updates)"
-            shift
-            ;;
-        --drop)
-            DROP_MODE="--drop"
-            echo "Modus: Drop (bestehende Tabellen loeschen)"
-            shift
-            ;;
-        *)
-            echo "Unbekannte Option: $1"
-            exit 1
-            ;;
-    esac
-done
-
-# Estimate time
-PBF_SIZE=$(stat -f%z "$PBF_FILE" 2>/dev/null || stat -c%s "$PBF_FILE")
-ESTIMATED_HOURS=$((PBF_SIZE / 1000000000 / 2))  # Rough estimate: ~2GB per hour
-
-echo ""
-echo -e "${YELLOW}ACHTUNG: Import kann 2-4 Stunden dauern!${NC}"
-echo "Geschaetzte Dauer: ~${ESTIMATED_HOURS}-$((ESTIMATED_HOURS * 2)) Stunden"
-echo ""
-read -p "Import starten? (j/N) " -n 1 -r
-echo ""
-
-if [[ ! $REPLY =~ ^[JjYy]$ ]]; then
-    echo "Import abgebrochen."
-    exit 1
-fi
-
-# Create style file
-STYLE_FILE="${DATA_DIR}/osm2pgsql.style"
-
-cat > "$STYLE_FILE" << 'EOF'
-# osm2pgsql style file for GeoEdu Service
-# Optimized for educational geography content
-
-# Common tags
-node,way   access        text         linear
-node,way   addr:housename text        linear
-node,way   addr:housenumber text      linear
-node,way   addr:interpolation text    linear
-node,way   admin_level   text         linear
-node,way   aerialway     text         linear
-node,way   amenity       text         polygon
-node,way   area          text         linear
-node,way   barrier       text         linear
-node,way   bicycle       text         linear
-node,way   boundary      text         linear
-node,way   bridge        text         linear
-node,way   building      text         polygon
-node,way   construction  text         linear
-node,way   covered       text         linear
-node,way   foot          text         linear
-node,way   highway       text         linear
-node,way   historic      text         polygon
-node,way   junction      text         linear
-node,way   landuse       text         polygon
-node,way   layer         text         linear
-node,way   leisure       text         polygon
-node,way   man_made      text         polygon
-node,way   military      text         polygon
-node,way   name          text         linear
-node,way   natural       text         polygon
-node,way   oneway        text         linear
-node,way   place         text         polygon
-node,way   power         text         polygon
-node,way   railway       text         linear
-node,way   ref           text         linear
-node,way   religion      text         linear
-node,way   route         text         linear
-node,way   service       text         linear
-node,way   shop          text         polygon
-node,way   sport         text         polygon
-node,way   surface       text         linear
-node,way   tourism       text         polygon
-node,way   tracktype     text         linear
-node,way   tunnel        text         linear
-node,way   water         text         polygon
-node,way   waterway      text         linear
-node,way   wetland       text         polygon
-node,way   wood          text         polygon
-node,way   z_order       int4         linear
-
-# Elevation data from OSM
-node,way   ele           text         linear
-
-# Name translations
-node,way   name:de       text         linear
-node,way   name:en       text         linear
-
-# Population for place rendering
-node,way   population    text         linear
-
-# Way area for polygon ordering
-way        way_area      real         linear
-EOF
-
-echo ""
-echo -e "${GREEN}[1/3] PostGIS Extension aktivieren...${NC}"
-
-PGPASSWORD="$DB_PASS" psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d "$DB_NAME" -c "CREATE EXTENSION IF NOT EXISTS postgis;"
-PGPASSWORD="$DB_PASS" psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d "$DB_NAME" -c "CREATE EXTENSION IF NOT EXISTS hstore;"
-
-echo ""
-echo -e "${GREEN}[2/3] OSM Daten importieren...${NC}"
-echo "       Dies dauert mehrere Stunden!"
-echo ""
-
-# Run osm2pgsql
-PGPASSWORD="$DB_PASS" osm2pgsql \
-    -H "$DB_HOST" \
-    -P "$DB_PORT" \
-    -U "$DB_USER" \
-    -d "$DB_NAME" \
-    -S "$STYLE_FILE" \
-    --cache 4000 \
-    --number-processes 4 \
-    $SLIM_MODE \
-    $DROP_MODE \
-    "$PBF_FILE"
-
-echo ""
-echo -e "${GREEN}[3/3] Indizes erstellen...${NC}"
-
-PGPASSWORD="$DB_PASS" psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d "$DB_NAME" << 'EOSQL'
-- Spatial indexes (if not created by osm2pgsql)
-CREATE INDEX IF NOT EXISTS planet_osm_point_way_idx ON planet_osm_point USING GIST (way);
-CREATE INDEX IF NOT EXISTS planet_osm_line_way_idx ON planet_osm_line USING GIST (way);
-CREATE INDEX IF NOT EXISTS planet_osm_polygon_way_idx ON planet_osm_polygon USING GIST (way);
-CREATE INDEX IF NOT EXISTS planet_osm_roads_way_idx ON planet_osm_roads USING GIST (way);
-
-- Name indexes for search
-CREATE INDEX IF NOT EXISTS planet_osm_point_name_idx ON planet_osm_point (name);
-CREATE INDEX IF NOT EXISTS planet_osm_polygon_name_idx ON planet_osm_polygon (name);
-
-- Vacuum analyze for query optimization
-VACUUM ANALYZE planet_osm_point;
-VACUUM ANALYZE planet_osm_line;
-VACUUM ANALYZE planet_osm_polygon;
-VACUUM ANALYZE planet_osm_roads;
-EOSQL
-
-echo ""
-echo -e "${GREEN}============================================${NC}"
-echo -e "${GREEN}Import abgeschlossen!${NC}"
-echo -e "${GREEN}============================================${NC}"
-echo ""
-echo "Tabellen erstellt:"
-echo "  - planet_osm_point"
-echo "  - planet_osm_line"
-echo "  - planet_osm_polygon"
-echo "  - planet_osm_roads"
-echo ""
-echo -e "${YELLOW}Naechster Schritt:${NC}"
-echo "  ./generate_tiles.sh  # PMTiles generieren"
--- a/geo-service/services/init.py
+++ b/geo-service/services/init.py
@@ -1,16 +0,0 @@
-"""
-GeoEdu Service - Business Logic Services
-"""
-from .tile_server import TileServerService
-from .dem_service import DEMService
-from .aoi_packager import AOIPackagerService
-from .osm_extractor import OSMExtractorService
-from .learning_generator import LearningGeneratorService
-
-__all__ = [
-    "TileServerService",
-    "DEMService",
-    "AOIPackagerService",
-    "OSMExtractorService",
-    "LearningGeneratorService",
-]
--- a/geo-service/services/aoi_packager.py
+++ b/geo-service/services/aoi_packager.py
@@ -1,420 +0,0 @@
-"""
-AOI Packager Service
-Creates Unity-compatible bundles from geographic areas
-"""
-import os
-import json
-import zipfile
-import uuid
-from typing import Optional, Tuple
-from datetime import datetime
-import math
-import structlog
-from shapely.geometry import shape, Polygon, mapping
-from shapely.ops import transform
-import pyproj
-
-from config import settings
-
-logger = structlog.get_logger(__name__)
-
-# Germany bounding box
-GERMANY_BOUNDS = Polygon([
-    (5.87, 47.27),
-    (15.04, 47.27),
-    (15.04, 55.06),
-    (5.87, 55.06),
-    (5.87, 47.27),
-])
-
-# AOI status storage (in production, use database)
-_aoi_storage = {}
-
-
-class AOIPackagerService:
-    """
-    Service for packaging geographic areas for Unity 3D rendering.
-
-    Creates bundles containing:
-    - Terrain heightmap
-    - OSM features (buildings, roads, water, etc.)
-    - Learning node positions
-    - Attribution information
-    """
-
-    def __init__(self):
-        self.bundle_dir = settings.bundle_dir
-        self.max_area_km2 = settings.max_aoi_size_km2
-
-    def calculate_area_km2(self, geojson: dict) -> float:
-        """
-        Calculate the area of a GeoJSON polygon in square kilometers.
-
-        Uses an equal-area projection for accurate measurement.
-        """
-        try:
-            geom = shape(geojson)
-
-            # Transform to equal-area projection (EPSG:3035 for Europe)
-            project = pyproj.Transformer.from_crs(
-                "EPSG:4326",  # WGS84
-                "EPSG:3035",  # ETRS89-LAEA
-                always_xy=True,
-            ).transform
-
-            geom_projected = transform(project, geom)
-            area_m2 = geom_projected.area
-            area_km2 = area_m2 / 1_000_000
-
-            return area_km2
-
-        except Exception as e:
-            logger.error("Error calculating area", error=str(e))
-            raise ValueError(f"Invalid polygon geometry: {str(e)}")
-
-    def is_within_germany(self, geojson: dict) -> bool:
-        """Check if a polygon is within Germany's bounds."""
-        try:
-            geom = shape(geojson)
-            return GERMANY_BOUNDS.contains(geom)
-        except Exception:
-            return False
-
-    def validate_polygon(self, geojson: dict) -> Tuple[bool, str]:
-        """
-        Validate a GeoJSON polygon.
-
-        Checks:
-        - Valid GeoJSON format
-        - Valid polygon geometry
-        - Not self-intersecting
-        """
-        try:
-            # Check type
-            if geojson.get("type") != "Polygon":
-                return False, "Geometry must be a Polygon"
-
-            # Check coordinates
-            coords = geojson.get("coordinates")
-            if not coords or not isinstance(coords, list):
-                return False, "Missing or invalid coordinates"
-
-            # Parse geometry
-            geom = shape(geojson)
-
-            # Check validity
-            if not geom.is_valid:
-                return False, "Invalid polygon geometry (possibly self-intersecting)"
-
-            # Check ring closure
-            outer_ring = coords[0]
-            if outer_ring[0] != outer_ring[-1]:
-                return False, "Polygon ring must be closed"
-
-            return True, "Valid"
-
-        except Exception as e:
-            return False, f"Error validating polygon: {str(e)}"
-
-    def estimate_bundle_size_mb(self, area_km2: float, quality: str) -> float:
-        """Estimate the bundle size based on area and quality."""
-        # Base size per km² in MB
-        base_sizes = {
-            "low": 10,
-            "medium": 25,
-            "high": 50,
-        }
-
-        base = base_sizes.get(quality, 25)
-        return round(area_km2 * base, 1)
-
-    async def process_aoi(
-        self,
-        aoi_id: str,
-        polygon: dict,
-        theme: str,
-        quality: str,
-    ):
-        """
-        Process an AOI and create the Unity bundle.
-
-        This runs as a background task.
-        """
-        logger.info("Processing AOI", aoi_id=aoi_id, theme=theme, quality=quality)
-
-        # Update status
-        _aoi_storage[aoi_id] = {
-            "status": "processing",
-            "polygon": polygon,
-            "theme": theme,
-            "quality": quality,
-            "area_km2": self.calculate_area_km2(polygon),
-            "created_at": datetime.utcnow().isoformat(),
-        }
-
-        try:
-            # Create bundle directory
-            bundle_path = os.path.join(self.bundle_dir, aoi_id)
-            os.makedirs(bundle_path, exist_ok=True)
-
-            # Generate terrain heightmap
-            await self._generate_terrain(aoi_id, polygon, quality)
-
-            # Extract OSM features
-            await self._extract_osm_features(aoi_id, polygon)
-
-            # Generate learning node positions
-            await self._generate_learning_positions(aoi_id, polygon, theme)
-
-            # Create attribution file
-            await self._create_attribution(aoi_id)
-
-            # Create manifest
-            await self._create_manifest(aoi_id, polygon, theme, quality)
-
-            # Create ZIP bundle
-            await self._create_zip_bundle(aoi_id)
-
-            # Update status
-            _aoi_storage[aoi_id]["status"] = "completed"
-            _aoi_storage[aoi_id]["completed_at"] = datetime.utcnow().isoformat()
-
-            logger.info("AOI processing complete", aoi_id=aoi_id)
-
-        except Exception as e:
-            logger.error("AOI processing failed", aoi_id=aoi_id, error=str(e))
-            _aoi_storage[aoi_id]["status"] = "failed"
-            _aoi_storage[aoi_id]["error"] = str(e)
-
-    async def _generate_terrain(self, aoi_id: str, polygon: dict, quality: str):
-        """Generate terrain heightmap for the AOI."""
-        from services.dem_service import DEMService
-
-        dem_service = DEMService()
-        bundle_path = os.path.join(self.bundle_dir, aoi_id)
-
-        # Get bounding box of polygon
-        geom = shape(polygon)
-        bounds = geom.bounds  # (minx, miny, maxx, maxy)
-
-        # Determine resolution based on quality
-        resolutions = {"low": 64, "medium": 256, "high": 512}
-        resolution = resolutions.get(quality, 256)
-
-        # Generate heightmap for bounds
-        from services.dem_service import tile_to_bounds
-
-        # For simplicity, generate a single heightmap image
-        # In production, this would be more sophisticated
-
-        heightmap_path = os.path.join(bundle_path, "terrain.heightmap.png")
-
-        # Save bounds info
-        terrain_info = {
-            "bounds": {
-                "west": bounds[0],
-                "south": bounds[1],
-                "east": bounds[2],
-                "north": bounds[3],
-            },
-            "resolution": resolution,
-            "heightmap_file": "terrain.heightmap.png",
-            "encoding": "terrain-rgb",
-        }
-
-        with open(os.path.join(bundle_path, "terrain.json"), "w") as f:
-            json.dump(terrain_info, f, indent=2)
-
-        logger.debug("Terrain generated", aoi_id=aoi_id, resolution=resolution)
-
-    async def _extract_osm_features(self, aoi_id: str, polygon: dict):
-        """Extract OSM features within the AOI."""
-        from services.osm_extractor import OSMExtractorService
-
-        extractor = OSMExtractorService()
-        bundle_path = os.path.join(self.bundle_dir, aoi_id)
-
-        # Extract features
-        features = await extractor.extract_features(polygon)
-
-        # Save to file
-        features_path = os.path.join(bundle_path, "osm_features.json")
-        with open(features_path, "w") as f:
-            json.dump(features, f, indent=2)
-
-        logger.debug("OSM features extracted", aoi_id=aoi_id, count=len(features.get("features", [])))
-
-    async def _generate_learning_positions(self, aoi_id: str, polygon: dict, theme: str):
-        """Generate suggested positions for learning nodes."""
-        geom = shape(polygon)
-        bounds = geom.bounds
-        centroid = geom.centroid
-
-        # Generate positions based on theme
-        # For now, create a grid of potential positions
-        positions = []
-
-        # Create a 3x3 grid of positions
-        for i in range(3):
-            for j in range(3):
-                lon = bounds[0] + (bounds[2] - bounds[0]) * (i + 0.5) / 3
-                lat = bounds[1] + (bounds[3] - bounds[1]) * (j + 0.5) / 3
-
-                # Check if point is within polygon
-                from shapely.geometry import Point
-                if geom.contains(Point(lon, lat)):
-                    positions.append({
-                        "id": str(uuid.uuid4()),
-                        "position": {"longitude": lon, "latitude": lat},
-                        "suggested_theme": theme,
-                        "status": "pending",
-                    })
-
-        bundle_path = os.path.join(self.bundle_dir, aoi_id)
-        positions_path = os.path.join(bundle_path, "learning_positions.json")
-
-        with open(positions_path, "w") as f:
-            json.dump({"positions": positions}, f, indent=2)
-
-        logger.debug("Learning positions generated", aoi_id=aoi_id, count=len(positions))
-
-    async def _create_attribution(self, aoi_id: str):
-        """Create attribution file with required license notices."""
-        attribution = {
-            "sources": [
-                {
-                    "name": "OpenStreetMap",
-                    "license": "Open Database License (ODbL) v1.0",
-                    "url": "https://www.openstreetmap.org/copyright",
-                    "attribution": "© OpenStreetMap contributors",
-                    "required": True,
-                },
-                {
-                    "name": "Copernicus DEM",
-                    "license": "Copernicus Data License",
-                    "url": "https://spacedata.copernicus.eu/",
-                    "attribution": "© Copernicus Service Information 2024",
-                    "required": True,
-                },
-            ],
-            "generated_at": datetime.utcnow().isoformat(),
-            "notice": "This data must be attributed according to the licenses above when used publicly.",
-        }
-
-        bundle_path = os.path.join(self.bundle_dir, aoi_id)
-        attribution_path = os.path.join(bundle_path, "attribution.json")
-
-        with open(attribution_path, "w") as f:
-            json.dump(attribution, f, indent=2)
-
-    async def _create_manifest(self, aoi_id: str, polygon: dict, theme: str, quality: str):
-        """Create Unity bundle manifest."""
-        geom = shape(polygon)
-        bounds = geom.bounds
-        centroid = geom.centroid
-
-        manifest = {
-            "version": "1.0.0",
-            "aoi_id": aoi_id,
-            "created_at": datetime.utcnow().isoformat(),
-            "bounds": {
-                "west": bounds[0],
-                "south": bounds[1],
-                "east": bounds[2],
-                "north": bounds[3],
-            },
-            "center": {
-                "longitude": centroid.x,
-                "latitude": centroid.y,
-            },
-            "area_km2": self.calculate_area_km2(polygon),
-            "theme": theme,
-            "quality": quality,
-            "assets": {
-                "terrain": {
-                    "file": "terrain.heightmap.png",
-                    "config": "terrain.json",
-                },
-                "osm_features": {
-                    "file": "osm_features.json",
-                },
-                "learning_positions": {
-                    "file": "learning_positions.json",
-                },
-                "attribution": {
-                    "file": "attribution.json",
-                },
-            },
-            "unity": {
-                "coordinate_system": "Unity (Y-up, left-handed)",
-                "scale": 1.0,  # 1 Unity unit = 1 meter
-                "terrain_resolution": {"low": 64, "medium": 256, "high": 512}[quality],
-            },
-        }
-
-        bundle_path = os.path.join(self.bundle_dir, aoi_id)
-        manifest_path = os.path.join(bundle_path, "manifest.json")
-
-        with open(manifest_path, "w") as f:
-            json.dump(manifest, f, indent=2)
-
-    async def _create_zip_bundle(self, aoi_id: str):
-        """Create ZIP archive of all bundle files."""
-        bundle_path = os.path.join(self.bundle_dir, aoi_id)
-        zip_path = os.path.join(bundle_path, "bundle.zip")
-
-        with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
-            for filename in os.listdir(bundle_path):
-                if filename != "bundle.zip":
-                    filepath = os.path.join(bundle_path, filename)
-                    zf.write(filepath, filename)
-
-        logger.debug("Bundle ZIP created", aoi_id=aoi_id, path=zip_path)
-
-    async def get_aoi_status(self, aoi_id: str) -> Optional[dict]:
-        """Get the status of an AOI."""
-        return _aoi_storage.get(aoi_id)
-
-    async def get_manifest(self, aoi_id: str) -> Optional[dict]:
-        """Get the manifest for a completed AOI."""
-        aoi_data = _aoi_storage.get(aoi_id)
-        if aoi_data is None or aoi_data.get("status") != "completed":
-            return None
-
-        manifest_path = os.path.join(self.bundle_dir, aoi_id, "manifest.json")
-        if not os.path.exists(manifest_path):
-            return None
-
-        with open(manifest_path) as f:
-            return json.load(f)
-
-    async def get_bundle_path(self, aoi_id: str) -> Optional[str]:
-        """Get the path to a completed bundle ZIP."""
-        aoi_data = _aoi_storage.get(aoi_id)
-        if aoi_data is None or aoi_data.get("status") != "completed":
-            return None
-
-        zip_path = os.path.join(self.bundle_dir, aoi_id, "bundle.zip")
-        if not os.path.exists(zip_path):
-            return None
-
-        return zip_path
-
-    async def delete_aoi(self, aoi_id: str) -> bool:
-        """Delete an AOI and its files."""
-        if aoi_id not in _aoi_storage:
-            return False
-
-        import shutil
-
-        bundle_path = os.path.join(self.bundle_dir, aoi_id)
-        if os.path.exists(bundle_path):
-            shutil.rmtree(bundle_path)
-
-        del _aoi_storage[aoi_id]
-        return True
-
-    async def generate_preview(self, aoi_id: str, width: int, height: int) -> Optional[bytes]:
-        """Generate a preview image of the AOI (stub)."""
-        # Would generate a preview combining terrain and OSM features
-        return None
--- a/geo-service/services/dem_service.py
+++ b/geo-service/services/dem_service.py
@@ -1,338 +0,0 @@
-"""
-DEM (Digital Elevation Model) Service
-Serves terrain data from Copernicus DEM GLO-30
-"""
-import os
-import math
-from typing import Optional, Tuple
-import struct
-import structlog
-import numpy as np
-from PIL import Image
-from io import BytesIO
-
-from config import settings
-
-logger = structlog.get_logger(__name__)
-
-# Germany bounding box
-GERMANY_BOUNDS = {
-    "west": 5.87,
-    "south": 47.27,
-    "east": 15.04,
-    "north": 55.06,
-}
-
-
-def lat_lon_to_tile(lat: float, lon: float, zoom: int) -> Tuple[int, int]:
-    """Convert latitude/longitude to tile coordinates."""
-    n = 2 ** zoom
-    x = int((lon + 180.0) / 360.0 * n)
-    lat_rad = math.radians(lat)
-    y = int((1.0 - math.asinh(math.tan(lat_rad)) / math.pi) / 2.0 * n)
-    return x, y
-
-
-def tile_to_bounds(z: int, x: int, y: int) -> Tuple[float, float, float, float]:
-    """Convert tile coordinates to bounding box (west, south, east, north)."""
-    n = 2 ** z
-    west = x / n * 360.0 - 180.0
-    east = (x + 1) / n * 360.0 - 180.0
-    north_rad = math.atan(math.sinh(math.pi * (1 - 2 * y / n)))
-    south_rad = math.atan(math.sinh(math.pi * (1 - 2 * (y + 1) / n)))
-    north = math.degrees(north_rad)
-    south = math.degrees(south_rad)
-    return west, south, east, north
-
-
-class DEMService:
-    """
-    Service for handling Digital Elevation Model data.
-
-    Uses Copernicus DEM GLO-30 (30m resolution) as the data source.
-    Generates terrain tiles in Mapbox Terrain-RGB format for MapLibre/Unity.
-    """
-
-    def __init__(self):
-        self.dem_dir = settings.dem_data_dir
-        self.tile_size = settings.terrain_tile_size
-        self._dem_cache = {}  # Cache loaded DEM files
-
-    def _get_dem_file_path(self, lat: int, lon: int) -> str:
-        """
-        Get the path to a Copernicus DEM file for the given coordinates.
-
-        Copernicus DEM files are named like: Copernicus_DSM_COG_10_N47_00_E008_00_DEM.tif
-        """
-        lat_prefix = "N" if lat >= 0 else "S"
-        lon_prefix = "E" if lon >= 0 else "W"
-
-        # Format: N47E008
-        filename = f"{lat_prefix}{abs(lat):02d}{lon_prefix}{abs(lon):03d}.tif"
-        return os.path.join(self.dem_dir, filename)
-
-    def _load_dem_tile(self, lat: int, lon: int) -> Optional[np.ndarray]:
-        """Load a DEM tile from disk."""
-        cache_key = f"{lat}_{lon}"
-        if cache_key in self._dem_cache:
-            return self._dem_cache[cache_key]
-
-        filepath = self._get_dem_file_path(lat, lon)
-
-        if not os.path.exists(filepath):
-            logger.debug("DEM file not found", path=filepath)
-            return None
-
-        try:
-            import rasterio
-
-            with rasterio.open(filepath) as src:
-                data = src.read(1)  # Read first band
-                self._dem_cache[cache_key] = data
-                return data
-
-        except ImportError:
-            logger.warning("rasterio not available, using fallback")
-            return self._load_dem_fallback(filepath)
-        except Exception as e:
-            logger.error("Error loading DEM", path=filepath, error=str(e))
-            return None
-
-    def _load_dem_fallback(self, filepath: str) -> Optional[np.ndarray]:
-        """Fallback DEM loader without rasterio (for development)."""
-        # In development, return synthetic terrain
-        return None
-
-    async def get_heightmap_tile(self, z: int, x: int, y: int) -> Optional[bytes]:
-        """
-        Generate a heightmap tile in Mapbox Terrain-RGB format.
-
-        The encoding stores elevation as: height = -10000 + ((R * 256 * 256 + G * 256 + B) * 0.1)
-        This allows for -10000m to +1677721.6m range with 0.1m precision.
-        """
-        # Get tile bounds
-        west, south, east, north = tile_to_bounds(z, x, y)
-
-        # Check if tile is within Germany
-        if east < GERMANY_BOUNDS["west"] or west > GERMANY_BOUNDS["east"]:
-            return None
-        if north < GERMANY_BOUNDS["south"] or south > GERMANY_BOUNDS["north"]:
-            return None
-
-        # Try to load elevation data for this tile
-        elevations = await self._get_elevations_for_bounds(west, south, east, north)
-
-        if elevations is None:
-            # No DEM data available - return placeholder or None
-            return self._generate_placeholder_heightmap()
-
-        # Convert to Terrain-RGB format
-        return self._encode_terrain_rgb(elevations)
-
-    async def _get_elevations_for_bounds(
-        self, west: float, south: float, east: float, north: float
-    ) -> Optional[np.ndarray]:
-        """Get elevation data for a bounding box."""
-        # Determine which DEM tiles we need
-        lat_min = int(math.floor(south))
-        lat_max = int(math.ceil(north))
-        lon_min = int(math.floor(west))
-        lon_max = int(math.ceil(east))
-
-        # Load all required DEM tiles
-        dem_tiles = []
-        for lat in range(lat_min, lat_max + 1):
-            for lon in range(lon_min, lon_max + 1):
-                tile = self._load_dem_tile(lat, lon)
-                if tile is not None:
-                    dem_tiles.append((lat, lon, tile))
-
-        if not dem_tiles:
-            return None
-
-        # Interpolate elevations for our tile grid
-        elevations = np.zeros((self.tile_size, self.tile_size), dtype=np.float32)
-
-        for py in range(self.tile_size):
-            for px in range(self.tile_size):
-                # Calculate lat/lon for this pixel
-                lon = west + (east - west) * px / self.tile_size
-                lat = north - (north - south) * py / self.tile_size
-
-                # Get elevation (simplified - would need proper interpolation)
-                elevation = self._sample_elevation(lat, lon, dem_tiles)
-                elevations[py, px] = elevation if elevation is not None else 0
-
-        return elevations
-
-    def _sample_elevation(
-        self, lat: float, lon: float, dem_tiles: list
-    ) -> Optional[float]:
-        """Sample elevation at a specific point from loaded DEM tiles."""
-        tile_lat = int(math.floor(lat))
-        tile_lon = int(math.floor(lon))
-
-        for t_lat, t_lon, data in dem_tiles:
-            if t_lat == tile_lat and t_lon == tile_lon:
-                # Calculate pixel position within tile
-                # Assuming 1 degree = 3600 pixels (1 arcsecond for 30m DEM)
-                rows, cols = data.shape
-                px = int((lon - tile_lon) * cols)
-                py = int((t_lat + 1 - lat) * rows)
-
-                px = max(0, min(cols - 1, px))
-                py = max(0, min(rows - 1, py))
-
-                return float(data[py, px])
-
-        return None
-
-    def _encode_terrain_rgb(self, elevations: np.ndarray) -> bytes:
-        """
-        Encode elevation data as Mapbox Terrain-RGB.
-
-        Format: height = -10000 + ((R * 256 * 256 + G * 256 + B) * 0.1)
-        """
-        # Convert elevation to RGB values
-        # encoded = (elevation + 10000) / 0.1
-        encoded = ((elevations + 10000) / 0.1).astype(np.uint32)
-
-        r = (encoded // (256 * 256)) % 256
-        g = (encoded // 256) % 256
-        b = encoded % 256
-
-        # Create RGB image
-        rgb = np.stack([r, g, b], axis=-1).astype(np.uint8)
-        img = Image.fromarray(rgb, mode="RGB")
-
-        # Save to bytes
-        buffer = BytesIO()
-        img.save(buffer, format="PNG")
-        return buffer.getvalue()
-
-    def _generate_placeholder_heightmap(self) -> bytes:
-        """Generate a flat placeholder heightmap (sea level)."""
-        # Sea level = 0m encoded as RGB
-        # encoded = (0 + 10000) / 0.1 = 100000
-        # R = 100000 // 65536 = 1
-        # G = (100000 // 256) % 256 = 134
-        # B = 100000 % 256 = 160
-
-        img = Image.new("RGB", (self.tile_size, self.tile_size), (1, 134, 160))
-        buffer = BytesIO()
-        img.save(buffer, format="PNG")
-        return buffer.getvalue()
-
-    async def get_hillshade_tile(
-        self, z: int, x: int, y: int, azimuth: float = 315, altitude: float = 45
-    ) -> Optional[bytes]:
-        """
-        Generate a hillshade tile for terrain visualization.
-
-        Args:
-            z, x, y: Tile coordinates
-            azimuth: Light direction in degrees (0=N, 90=E, 180=S, 270=W)
-            altitude: Light altitude in degrees above horizon
-        """
-        west, south, east, north = tile_to_bounds(z, x, y)
-        elevations = await self._get_elevations_for_bounds(west, south, east, north)
-
-        if elevations is None:
-            return None
-
-        # Calculate hillshade using numpy
-        hillshade = self._calculate_hillshade(elevations, azimuth, altitude)
-
-        # Create grayscale image
-        img = Image.fromarray((hillshade * 255).astype(np.uint8), mode="L")
-
-        buffer = BytesIO()
-        img.save(buffer, format="PNG")
-        return buffer.getvalue()
-
-    def _calculate_hillshade(
-        self, dem: np.ndarray, azimuth: float, altitude: float
-    ) -> np.ndarray:
-        """Calculate hillshade from DEM array."""
-        # Convert angles to radians
-        azimuth_rad = math.radians(360 - azimuth + 90)
-        altitude_rad = math.radians(altitude)
-
-        # Calculate gradient
-        dy, dx = np.gradient(dem)
-
-        # Calculate slope and aspect
-        slope = np.arctan(np.sqrt(dx**2 + dy**2))
-        aspect = np.arctan2(-dy, dx)
-
-        # Calculate hillshade
-        hillshade = (
-            np.sin(altitude_rad) * np.cos(slope)
-            + np.cos(altitude_rad) * np.sin(slope) * np.cos(azimuth_rad - aspect)
-        )
-
-        # Normalize to 0-1
-        hillshade = np.clip(hillshade, 0, 1)
-
-        return hillshade
-
-    async def get_contour_tile(
-        self, z: int, x: int, y: int, interval: int = 20
-    ) -> Optional[bytes]:
-        """Generate contour lines as vector tile (stub - requires more complex implementation)."""
-        # This would require generating contour lines from DEM and encoding as MVT
-        # For now, return None
-        logger.warning("Contour tiles not yet implemented")
-        return None
-
-    async def get_elevation(self, lat: float, lon: float) -> Optional[float]:
-        """Get elevation at a specific point."""
-        tile_lat = int(math.floor(lat))
-        tile_lon = int(math.floor(lon))
-
-        dem_data = self._load_dem_tile(tile_lat, tile_lon)
-        if dem_data is None:
-            return None
-
-        return self._sample_elevation(lat, lon, [(tile_lat, tile_lon, dem_data)])
-
-    async def get_elevation_profile(
-        self, coordinates: list[list[float]], samples: int = 100
-    ) -> list[dict]:
-        """Get elevation profile along a path."""
-        from shapely.geometry import LineString
-        from shapely.ops import substring
-
-        # Create line from coordinates
-        line = LineString(coordinates)
-        total_length = line.length
-
-        # Sample points along line
-        profile = []
-        for i in range(samples):
-            fraction = i / (samples - 1)
-            point = line.interpolate(fraction, normalized=True)
-
-            elevation = await self.get_elevation(point.y, point.x)
-
-            profile.append({
-                "distance_m": fraction * total_length * 111320,  # Approximate meters
-                "longitude": point.x,
-                "latitude": point.y,
-                "elevation_m": elevation,
-            })
-
-        return profile
-
-    async def get_metadata(self) -> dict:
-        """Get metadata about available DEM data."""
-        dem_files = []
-        if os.path.exists(self.dem_dir):
-            dem_files = [f for f in os.listdir(self.dem_dir) if f.endswith(".tif")]
-
-        return {
-            "data_available": len(dem_files) > 0,
-            "tiles_generated": len(dem_files),
-            "resolution_m": 30,
-            "source": "Copernicus DEM GLO-30",
-        }
--- a/geo-service/services/learning_generator.py
+++ b/geo-service/services/learning_generator.py
@@ -1,355 +0,0 @@
-"""
-Learning Generator Service
-Generates educational content for geographic areas using LLM
-"""
-import os
-import json
-import uuid
-from typing import Optional
-import structlog
-import httpx
-
-from config import settings
-from models.learning_node import LearningNode, LearningTheme, NodeType
-
-logger = structlog.get_logger(__name__)
-
-# In-memory storage for learning nodes (use database in production)
-_learning_nodes = {}
-
-
-class LearningGeneratorService:
-    """
-    Service for generating educational learning nodes using Ollama LLM.
-
-    Generates themed educational content based on geographic features
-    and didactic principles.
-    """
-
-    def __init__(self):
-        self.ollama_url = settings.ollama_base_url
-        self.model = settings.ollama_model
-        self.timeout = settings.ollama_timeout
-
-    async def generate_nodes(
-        self,
-        aoi_id: str,
-        theme: LearningTheme,
-        difficulty: str,
-        node_count: int,
-        grade_level: Optional[str] = None,
-        language: str = "de",
-    ) -> list[LearningNode]:
-        """
-        Generate learning nodes for an AOI.
-
-        Uses the Ollama LLM to create educational content appropriate
-        for the theme, difficulty, and grade level.
-        """
-        # Get AOI information
-        aoi_info = await self._get_aoi_info(aoi_id)
-        if aoi_info is None:
-            raise FileNotFoundError(f"AOI {aoi_id} not found")
-
-        # Build prompt for LLM
-        prompt = self._build_generation_prompt(
-            aoi_info=aoi_info,
-            theme=theme,
-            difficulty=difficulty,
-            node_count=node_count,
-            grade_level=grade_level,
-            language=language,
-        )
-
-        # Call Ollama
-        try:
-            response = await self._call_ollama(prompt)
-            nodes = self._parse_llm_response(response, aoi_id, theme)
-        except ConnectionError:
-            logger.warning("Ollama not available, using mock data")
-            nodes = self._generate_mock_nodes(aoi_id, theme, difficulty, node_count)
-
-        # Store nodes
-        if aoi_id not in _learning_nodes:
-            _learning_nodes[aoi_id] = []
-        _learning_nodes[aoi_id].extend(nodes)
-
-        return nodes
-
-    async def _get_aoi_info(self, aoi_id: str) -> Optional[dict]:
-        """Get information about an AOI from its manifest."""
-        manifest_path = os.path.join(settings.bundle_dir, aoi_id, "manifest.json")
-
-        if os.path.exists(manifest_path):
-            with open(manifest_path) as f:
-                return json.load(f)
-
-        # Check in-memory storage
-        from services.aoi_packager import _aoi_storage
-        return _aoi_storage.get(aoi_id)
-
-    def _build_generation_prompt(
-        self,
-        aoi_info: dict,
-        theme: LearningTheme,
-        difficulty: str,
-        node_count: int,
-        grade_level: Optional[str],
-        language: str,
-    ) -> str:
-        """Build a prompt for the LLM to generate learning nodes."""
-        theme_descriptions = {
-            LearningTheme.TOPOGRAPHIE: "Landschaftsformen, Höhen und Geländemerkmale",
-            LearningTheme.LANDNUTZUNG: "Siedlungen, Landwirtschaft und Flächennutzung",
-            LearningTheme.ORIENTIERUNG: "Kartenlesen, Kompass und Navigation",
-            LearningTheme.GEOLOGIE: "Gesteinsarten und geologische Formationen",
-            LearningTheme.HYDROLOGIE: "Gewässer, Einzugsgebiete und Wasserkreislauf",
-            LearningTheme.VEGETATION: "Pflanzengemeinschaften und Klimazonen",
-        }
-
-        difficulty_descriptions = {
-            "leicht": "Grundlegende Beobachtungen und einfache Fakten",
-            "mittel": "Verknüpfung von Zusammenhängen und Vergleiche",
-            "schwer": "Analyse, Transfer und kritisches Denken",
-        }
-
-        bounds = aoi_info.get("bounds", {})
-        center = aoi_info.get("center", {})
-
-        prompt = f"""Du bist ein Erdkunde-Didaktiker und erstellst Lernstationen für eine interaktive 3D-Lernwelt.
-
-GEBIET:
- Zentrum: {center.get('latitude', 0):.4f}°N, {center.get('longitude', 0):.4f}°E
- Fläche: ca. {aoi_info.get('area_km2', 0):.2f} km²
- Grenzen: West {bounds.get('west', 0):.4f}°, Süd {bounds.get('south', 0):.4f}°, Ost {bounds.get('east', 0):.4f}°, Nord {bounds.get('north', 0):.4f}°
-
-THEMA: {theme.value} - {theme_descriptions.get(theme, '')}
-
-SCHWIERIGKEITSGRAD: {difficulty} - {difficulty_descriptions.get(difficulty, '')}
-
-ZIELGRUPPE: {grade_level if grade_level else 'Allgemein (Klasse 5-10)'}
-
-AUFGABE:
-Erstelle {node_count} Lernstationen im JSON-Format. Jede Station soll:
-1. Eine geografische Position innerhalb des Gebiets haben
-2. Eine Lernfrage oder Aufgabe enthalten
-3. Hinweise zur Lösung bieten
-4. Die richtige Antwort mit Erklärung enthalten
-
-FORMAT (JSON-Array):
-[
-  {{
-    "title": "Titel der Station",
-    "position": {{"latitude": 0.0, "longitude": 0.0}},
-    "question": "Die Lernfrage",
-    "hints": ["Hinweis 1", "Hinweis 2"],
-    "answer": "Die Antwort",
-    "explanation": "Didaktische Erklärung",
-    "node_type": "question|observation|exploration",
-    "points": 10
-  }}
-]
-
-WICHTIG:
- Positionen müssen innerhalb der Gebietsgrenzen liegen
- Fragen sollen zum Thema {theme.value} passen
- Sprache: {"Deutsch" if language == "de" else "English"}
- Altersgerechte Formulierungen verwenden
-
-Antworte NUR mit dem JSON-Array, ohne weitere Erklärungen."""
-
-        return prompt
-
-    async def _call_ollama(self, prompt: str) -> str:
-        """Call Ollama API to generate content."""
-        try:
-            async with httpx.AsyncClient(timeout=self.timeout) as client:
-                response = await client.post(
-                    f"{self.ollama_url}/api/generate",
-                    json={
-                        "model": self.model,
-                        "prompt": prompt,
-                        "stream": False,
-                        "options": {
-                            "temperature": 0.7,
-                            "top_p": 0.9,
-                        },
-                    },
-                )
-
-                if response.status_code != 200:
-                    raise ConnectionError(f"Ollama returned {response.status_code}")
-
-                result = response.json()
-                return result.get("response", "")
-
-        except httpx.ConnectError:
-            raise ConnectionError("Cannot connect to Ollama")
-        except Exception as e:
-            logger.error("Ollama API error", error=str(e))
-            raise ConnectionError(f"Ollama error: {str(e)}")
-
-    def _parse_llm_response(
-        self, response: str, aoi_id: str, theme: LearningTheme
-    ) -> list[LearningNode]:
-        """Parse LLM response into LearningNode objects."""
-        try:
-            # Find JSON array in response
-            start = response.find("[")
-            end = response.rfind("]") + 1
-
-            if start == -1 or end == 0:
-                raise ValueError("No JSON array found in response")
-
-            json_str = response[start:end]
-            data = json.loads(json_str)
-
-            nodes = []
-            for item in data:
-                node = LearningNode(
-                    id=str(uuid.uuid4()),
-                    aoi_id=aoi_id,
-                    title=item.get("title", "Unbenannte Station"),
-                    theme=theme,
-                    position={
-                        "latitude": item.get("position", {}).get("latitude", 0),
-                        "longitude": item.get("position", {}).get("longitude", 0),
-                    },
-                    question=item.get("question", ""),
-                    hints=item.get("hints", []),
-                    answer=item.get("answer", ""),
-                    explanation=item.get("explanation", ""),
-                    node_type=NodeType(item.get("node_type", "question")),
-                    points=item.get("points", 10),
-                    approved=False,
-                )
-                nodes.append(node)
-
-            return nodes
-
-        except (json.JSONDecodeError, ValueError) as e:
-            logger.error("Failed to parse LLM response", error=str(e))
-            return []
-
-    def _generate_mock_nodes(
-        self,
-        aoi_id: str,
-        theme: LearningTheme,
-        difficulty: str,
-        node_count: int,
-    ) -> list[LearningNode]:
-        """Generate mock learning nodes for development."""
-        mock_questions = {
-            LearningTheme.TOPOGRAPHIE: [
-                ("Höhenbestimmung", "Schätze die Höhe dieses Punktes.", "Ca. 500m über NN"),
-                ("Hangneigung", "Beschreibe die Steilheit des Hanges.", "Mäßig steil, ca. 15-20°"),
-                ("Talform", "Welche Form hat dieses Tal?", "V-förmiges Erosionstal"),
-            ],
-            LearningTheme.LANDNUTZUNG: [
-                ("Gebäudetypen", "Welche Gebäude siehst du hier?", "Wohnhäuser und landwirtschaftliche Gebäude"),
-                ("Flächennutzung", "Wie wird das Land genutzt?", "Landwirtschaft und Siedlung"),
-                ("Infrastruktur", "Welche Verkehrswege erkennst du?", "Straße und Feldweg"),
-            ],
-            LearningTheme.ORIENTIERUNG: [
-                ("Himmelsrichtung", "In welche Richtung fließt der Bach?", "Nach Nordwesten"),
-                ("Entfernung", "Wie weit ist es bis zum Waldrand?", "Etwa 200 Meter"),
-                ("Wegbeschreibung", "Beschreibe den Weg zum Aussichtspunkt.", "Nordöstlich, bergauf"),
-            ],
-        }
-
-        questions = mock_questions.get(theme, mock_questions[LearningTheme.TOPOGRAPHIE])
-        nodes = []
-
-        for i in range(min(node_count, len(questions))):
-            title, question, answer = questions[i]
-            nodes.append(LearningNode(
-                id=str(uuid.uuid4()),
-                aoi_id=aoi_id,
-                title=title,
-                theme=theme,
-                position={"latitude": 47.7 + i * 0.001, "longitude": 9.19 + i * 0.001},
-                question=question,
-                hints=[f"Hinweis {j + 1}" for j in range(2)],
-                answer=answer,
-                explanation=f"Diese Aufgabe trainiert die Beobachtung von {theme.value}.",
-                node_type=NodeType.QUESTION,
-                points=10,
-                approved=False,
-            ))
-
-        return nodes
-
-    async def get_nodes_for_aoi(
-        self, aoi_id: str, theme: Optional[LearningTheme] = None
-    ) -> Optional[list[LearningNode]]:
-        """Get all learning nodes for an AOI."""
-        nodes = _learning_nodes.get(aoi_id)
-
-        if nodes is None:
-            return None
-
-        if theme is not None:
-            nodes = [n for n in nodes if n.theme == theme]
-
-        return nodes
-
-    async def update_node(
-        self, aoi_id: str, node_id: str, node_update: LearningNode
-    ) -> bool:
-        """Update a learning node."""
-        nodes = _learning_nodes.get(aoi_id)
-        if nodes is None:
-            return False
-
-        for i, node in enumerate(nodes):
-            if node.id == node_id:
-                _learning_nodes[aoi_id][i] = node_update
-                return True
-
-        return False
-
-    async def delete_node(self, aoi_id: str, node_id: str) -> bool:
-        """Delete a learning node."""
-        nodes = _learning_nodes.get(aoi_id)
-        if nodes is None:
-            return False
-
-        for i, node in enumerate(nodes):
-            if node.id == node_id:
-                del _learning_nodes[aoi_id][i]
-                return True
-
-        return False
-
-    async def approve_node(self, aoi_id: str, node_id: str) -> bool:
-        """Approve a learning node for student use."""
-        nodes = _learning_nodes.get(aoi_id)
-        if nodes is None:
-            return False
-
-        for node in nodes:
-            if node.id == node_id:
-                node.approved = True
-                return True
-
-        return False
-
-    async def get_statistics(self) -> dict:
-        """Get statistics about learning node usage."""
-        total = 0
-        by_theme = {}
-        by_difficulty = {}
-
-        for aoi_nodes in _learning_nodes.values():
-            for node in aoi_nodes:
-                total += 1
-                theme = node.theme.value
-                by_theme[theme] = by_theme.get(theme, 0) + 1
-
-        return {
-            "total_nodes": total,
-            "by_theme": by_theme,
-            "by_difficulty": by_difficulty,
-            "avg_per_aoi": total / len(_learning_nodes) if _learning_nodes else 0,
-            "popular_theme": max(by_theme, key=by_theme.get) if by_theme else "topographie",
-        }
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Sharang Parnerkar	b697963186	fix: use Alpine-compatible addgroup/adduser flags in Dockerfiles Some checks failed Deploy to Coolify / deploy (push) Has been cancelled Details Replace --system/--gid/--uid (Debian syntax) with -S/-g/-u (BusyBox/Alpine). Coolify ARG injection causes exit code 255 with Debian-style flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-06 22:38:31 +01:00
Sharang Parnerkar	ef6237ffdf	refactor(coolify): externalize postgres, qdrant, S3 Some checks failed Deploy to Coolify / deploy (push) Has been cancelled Details - Replace bp-core-postgres with POSTGRES_HOST env var - Replace bp-core-qdrant with QDRANT_URL env var - Replace bp-core-minio with S3_ENDPOINT/S3_ACCESS_KEY/S3_SECRET_KEY Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-03 09:23:32 +01:00
Sharang Parnerkar	41a8f3b183	feat: add Coolify deployment configuration Some checks failed Deploy to Coolify / deploy (push) Has been cancelled Details Add docker-compose.coolify.yml (8 services), .env.coolify.example, and Gitea Action workflow for Coolify API deployment. Removes core-health-check, paddleocr, transcription-worker, agent-core, drive, and docs. Adds Traefik labels for *.breakpilot.ai domain routing with Let's Encrypt SSL. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 10:43:15 +01:00
Benjamin Boenisch	414e0f5ec0	feat: edu-search-service migriert, voice-service/geo-service entfernt All checks were successful CI / go-lint (push) Has been skipped Details CI / python-lint (push) Has been skipped Details CI / nodejs-lint (push) Has been skipped Details CI / test-go-school (push) Successful in 28s Details CI / test-go-edu-search (push) Successful in 27s Details CI / test-python-klausur (push) Successful in 1m45s Details CI / test-python-agent-core (push) Successful in 16s Details CI / test-nodejs-website (push) Successful in 21s Details - edu-search-service von breakpilot-pwa nach breakpilot-lehrer kopiert (ohne vendor) - opensearch + edu-search-service in docker-compose.yml hinzugefuegt - voice-service aus docker-compose.yml entfernt (jetzt in breakpilot-core) - geo-service aus docker-compose.yml entfernt (nicht mehr benoetigt) - CI/CD: edu-search-service zu Gitea Actions und Woodpecker hinzugefuegt (Go lint, test mit go mod download, build, SBOM) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-15 18:36:38 +01:00
Benjamin Boenisch	d4e1d6bab6	fix: correct gradeToOberstufenPoints formula for grades < 2.0 All checks were successful CI / go-lint (push) Has been skipped Details CI / python-lint (push) Has been skipped Details CI / nodejs-lint (push) Has been skipped Details CI / test-go-school (push) Successful in 26s Details CI / test-python-klausur (push) Successful in 1m44s Details CI / test-python-agent-core (push) Successful in 16s Details CI / test-nodejs-website (push) Successful in 15s Details The formula int(17 - grade*3) was off by one for grades 1.0 and 1.3 due to integer truncation. Added +1 adjustment for grades < 2.0 to match the intended mapping (1.0=15, 1.3=14). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-15 17:32:24 +01:00
Benjamin Boenisch	ccff37f91b	fix(ci): replace actions/checkout with manual git clone Some checks failed CI / go-lint (push) Has been skipped Details CI / python-lint (push) Has been skipped Details CI / nodejs-lint (push) Has been skipped Details CI / test-go-school (push) Failing after 34s Details CI / test-python-klausur (push) Successful in 1m50s Details CI / test-python-agent-core (push) Successful in 17s Details CI / test-nodejs-website (push) Successful in 18s Details The act_runner cannot create /home/act_runner cache dir inside container images. Replace actions/checkout@v4 with manual git clone using GITHUB_SERVER_URL and GITHUB_REPOSITORY env vars. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>	2026-02-15 16:58:30 +01:00
Benjamin Boenisch	ce32850b09	fix(ci): use docker runner label instead of ubuntu-latest Some checks failed CI / go-lint (push) Has been skipped Details CI / python-lint (push) Has been skipped Details CI / nodejs-lint (push) Has been skipped Details CI / test-go-school (push) Failing after 1s Details CI / test-python-klausur (push) Failing after 1s Details CI / test-python-agent-core (push) Failing after 1s Details CI / test-nodejs-website (push) Failing after 6s Details The Gitea Actions runner on meghsakha uses label "docker". Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>	2026-02-15 16:53:32 +01:00
Benjamin Boenisch	50d0e65ef1	ci: add Gitea Actions workflow for external CI Some checks failed CI / go-lint (push) Has been cancelled Details CI / python-lint (push) Has been cancelled Details CI / nodejs-lint (push) Has been cancelled Details CI / test-go-school (push) Has been cancelled Details CI / test-python-klausur (push) Has been cancelled Details CI / test-python-agent-core (push) Has been cancelled Details CI / test-nodejs-website (push) Has been cancelled Details Adds .gitea/workflows/ci.yaml with lint and test jobs. Runs on gitea.meghsakha.com with Gitea Actions runner. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>	2026-02-15 16:39:00 +01:00
Benjamin Boenisch	4fd4b08f75	remove: geo-service komplett entfernt Der Geo-Service (DSGVO-konforme Geografie-Lernplattform) war ein nie genutzter Prototyp und wird nicht mehr benoetigt. Entfernt aus: - Quellcode (geo-service/) - CI/CD Pipeline (test, build, sbom, deploy steps) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>	2026-02-15 16:15:29 +01:00