breakpilot-compliance/backend-compliance/compliance/README_AI.md

Compliance AI Integration - Quick Start

Schnellstart (5 Minuten)

1. Environment Variables setzen

# In backend/.env
COMPLIANCE_LLM_PROVIDER=mock  # Für Testing ohne API-Key
# ODER
COMPLIANCE_LLM_PROVIDER=anthropic
ANTHROPIC_API_KEY=sk-ant-...

2. Backend starten

cd backend
docker-compose up -d
# ODER
uvicorn main:app --reload

3. Datenbank seeden

# Requirements und Module laden
curl -X POST http://localhost:8000/api/v1/compliance/seed \
  -H "Content-Type: application/json" \
  -d '{"force": false}'

curl -X POST http://localhost:8000/api/v1/compliance/modules/seed \
  -H "Content-Type: application/json" \
  -d '{"force": false}'

4. AI-Features testen

# Test-Script ausfuhren
python backend/scripts/test_compliance_ai_endpoints.py

API Endpoints

Alle Endpoints unter: http://localhost:8000/api/v1/compliance/ai/

1. Status prufen

curl http://localhost:8000/api/v1/compliance/ai/status

2. Requirement interpretieren

curl -X POST http://localhost:8000/api/v1/compliance/ai/interpret \
  -H "Content-Type: application/json" \
  -d '{
    "requirement_id": "YOUR_REQUIREMENT_ID"
  }'

3. Controls vorschlagen

curl -X POST http://localhost:8000/api/v1/compliance/ai/suggest-controls \
  -H "Content-Type: application/json" \
  -d '{
    "requirement_id": "YOUR_REQUIREMENT_ID"
  }'

4. Modul-Risiko bewerten

curl -X POST http://localhost:8000/api/v1/compliance/ai/assess-risk \
  -H "Content-Type: application/json" \
  -d '{
    "module_id": "consent-service"
  }'

5. Gap-Analyse

curl -X POST http://localhost:8000/api/v1/compliance/ai/gap-analysis \
  -H "Content-Type: application/json" \
  -d '{
    "requirement_id": "YOUR_REQUIREMENT_ID"
  }'

6. Batch-Interpretation

curl -X POST http://localhost:8000/api/v1/compliance/ai/batch-interpret \
  -H "Content-Type: application/json" \
  -d '{
    "requirement_ids": ["id1", "id2"],
    "rate_limit": 1.0
  }'

Provider-Konfiguration

Option 1: Mock (Testing)

export COMPLIANCE_LLM_PROVIDER=mock

Vorteile:

• Keine API-Keys erforderlich
• Schnell
• Deterministisch

Nachteile:

• Keine echten AI-Antworten

Option 2: Anthropic Claude (Empfohlen)

export COMPLIANCE_LLM_PROVIDER=anthropic
export ANTHROPIC_API_KEY=sk-ant-...
export ANTHROPIC_MODEL=claude-sonnet-4-20250514

Vorteile:

• Beste Qualitat
• Zuverlassig
• Breakpilot-optimiert

Nachteile:

• API-Kosten (~$3 per 1M input tokens)

Option 3: Self-Hosted (Ollama/vLLM)

export COMPLIANCE_LLM_PROVIDER=self_hosted
export SELF_HOSTED_LLM_URL=http://localhost:11434
export SELF_HOSTED_LLM_MODEL=llama3.1:8b

Vorteile:

• Kostenlos
• Privacy (on-premise)
• Keine Rate-Limits

Nachteile:

• Geringere Qualitat als Claude
• Benotigt GPU/CPU-Ressourcen

Beispiel-Response

Interpretation

{
  "requirement_id": "req-123",
  "summary": "Art. 32 DSGVO verlangt angemessene technische Maßnahmen zur Datensicherheit.",
  "applicability": "Gilt für alle Breakpilot-Module die personenbezogene Daten verarbeiten.",
  "technical_measures": [
    "Verschlüsselung personenbezogener Daten (AES-256)",
    "TLS 1.3 für Datenübertragung",
    "Regelmäßige Sicherheitsaudits",
    "Zugriffskontrolle mit IAM"
  ],
  "affected_modules": [
    "consent-service",
    "klausur-service",
    "backend"
  ],
  "risk_level": "high",
  "implementation_hints": [
    "SOPS mit Age-Keys für Secret-Management",
    "PostgreSQL transparent encryption",
    "Nginx TLS-Konfiguration prüfen"
  ],
  "confidence_score": 0.85,
  "error": null
}

Control-Suggestion

{
  "requirement_id": "req-123",
  "suggestions": [
    {
      "control_id": "PRIV-042",
      "domain": "priv",
      "title": "Verschlüsselung personenbezogener Daten",
      "description": "Alle personenbezogenen Daten müssen verschlüsselt gespeichert werden",
      "pass_criteria": "100% der PII in PostgreSQL sind AES-256 verschlüsselt",
      "implementation_guidance": "Verwende SOPS mit Age-Keys für Secrets. Aktiviere PostgreSQL transparent data encryption.",
      "is_automated": true,
      "automation_tool": "SOPS",
      "priority": "high",
      "confidence_score": 0.9
    }
  ]
}

Troubleshooting

Problem: "AI provider is not available"

Lösung:

# Prüfe Status
curl http://localhost:8000/api/v1/compliance/ai/status

# Prüfe Environment Variables
echo $COMPLIANCE_LLM_PROVIDER
echo $ANTHROPIC_API_KEY

# Fallback auf Mock
export COMPLIANCE_LLM_PROVIDER=mock

Problem: "Requirement not found"

Lösung:

# Datenbank seeden
curl -X POST http://localhost:8000/api/v1/compliance/seed \
  -d '{"force": false}'

# Requirements auflisten
curl http://localhost:8000/api/v1/compliance/requirements

Problem: Timeout bei Anthropic

Lösung:

# Timeout erhöhen
export COMPLIANCE_LLM_TIMEOUT=120.0

# Oder Mock-Provider verwenden
export COMPLIANCE_LLM_PROVIDER=mock

Unit Tests

cd backend

# Alle Tests
pytest tests/test_compliance_ai.py -v

# Nur Mock-Tests
pytest tests/test_compliance_ai.py::TestMockProvider -v

# Integration Tests (benötigt API-Key)
pytest tests/test_compliance_ai.py -v --integration

Weitere Dokumentation

• Vollständige Dokumentation: backend/docs/compliance_ai_integration.md
• API Schemas: backend/compliance/api/schemas.py
• LLM Provider: backend/compliance/services/llm_provider.py
• AI Assistant: backend/compliance/services/ai_compliance_assistant.py

Support

Bei Problemen:

1. Prüfe /api/v1/compliance/ai/status
2. Prüfe Logs: docker logs breakpilot-backend
3. Teste mit Mock: COMPLIANCE_LLM_PROVIDER=mock
4. Siehe: backend/docs/compliance_ai_integration.md