feat: Phase 3 — RAG-Anbindung fuer alle 18 Dokumenttypen + Vendor Contract Review

Migrate queryRAG from klausur-service GET to bp-core-rag-service POST with
multi-collection support. Each of the 18 ScopeDocumentType now gets a
type-specific RAG collection and optimized search query instead of the
generic fallback. Vendor-compliance contract review now uses LLM + RAG
for real analysis with mock fallback on error.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

admin-compliance/app/api/sdk/drafting-engine/draft/route.ts

@@ -32,6 +32,7 @@ import { terminologyToPromptString, styleContractToPromptString } from '@/lib/sd
 import { executeRepairLoop, type ProseBlockOutput, type RepairAudit } from '@/lib/sdk/drafting-engine/prose-validator'
 import { ProseCacheManager, computeChecksumSync, type CacheKeyParams } from '@/lib/sdk/drafting-engine/cache'
 import { queryRAG } from '@/lib/sdk/drafting-engine/rag-query'
+import { DOCUMENT_RAG_CONFIG } from '@/lib/sdk/drafting-engine/rag-config'
 
 // ============================================================================
 // Shared State
@@ -104,11 +105,9 @@ async function handleV1Draft(body: Record<string, unknown>): Promise<NextRespons
     }, { status: 403 })
   }
 
-  // RAG: Fetch relevant legal context
-  const ragQuery = documentType === 'dsfa'
-    ? 'Datenschutz-Folgenabschaetzung Art. 35 DSGVO Risikobewertung'
-    : `${documentType} DSGVO Compliance Anforderungen`
-  const ragContext = await queryRAG(ragQuery)
+  // RAG: Fetch relevant legal context (config-based)
+  const ragCfg = DOCUMENT_RAG_CONFIG[documentType]
+  const ragContext = await queryRAG(ragCfg.query, 3, ragCfg.collection)
 
   let v1SystemPrompt = V1_SYSTEM_PROMPT
   if (ragContext) {
@@ -380,11 +379,9 @@ async function handleV2Draft(body: Record<string, unknown>): Promise<NextRespons
   // Compute prompt hash for audit
   const promptHash = computeChecksumSync({ factsString, tagsString, termsString, styleString, disallowedString })
 
-  // Step 5b: RAG Legal Context
-  const v2RagQuery = documentType === 'dsfa'
-    ? 'DSFA Art. 35 DSGVO Risikobewertung Massnahmen Datenschutz-Folgenabschaetzung'
-    : `${documentType} DSGVO Compliance`
-  const v2RagContext = await queryRAG(v2RagQuery)
+  // Step 5b: RAG Legal Context (config-based)
+  const v2RagCfg = DOCUMENT_RAG_CONFIG[documentType]
+  const v2RagContext = await queryRAG(v2RagCfg.query, 3, v2RagCfg.collection)
 
   // Step 6: Generate Prose Blocks (with cache + repair loop)
   const proseBlocks = DOCUMENT_PROSE_BLOCKS[documentType] || DOCUMENT_PROSE_BLOCKS.tom

admin-compliance/app/api/sdk/v1/vendor-compliance/contracts/[id]/review/route.ts

@@ -4,11 +4,18 @@ import {
   Finding,
   CONTRACT_REVIEW_SYSTEM_PROMPT,
 } from '@/lib/sdk/vendor-compliance'
+import { queryRAG } from '@/lib/sdk/drafting-engine/rag-query'
+import { transformAnalysisResponse } from '@/lib/sdk/vendor-compliance/contract-review/analyzer'
+
+const OLLAMA_URL = process.env.OLLAMA_URL || 'http://host.docker.internal:11434'
+const LLM_MODEL = process.env.COMPLIANCE_LLM_MODEL || 'qwen2.5vl:32b'
 
 /**
  * POST /api/sdk/v1/vendor-compliance/contracts/[id]/review
  *
- * Starts the LLM-based contract review process
+ * Starts the LLM-based contract review process.
+ * If documentText is provided, runs LLM analysis with RAG context.
+ * Falls back to mock findings on LLM error or missing documentText.
  */
 export async function POST(
   request: NextRequest,
@@ -16,15 +23,84 @@ export async function POST(
 ) {
   try {
     const { id: contractId } = await params
+    const body = await request.json().catch(() => ({}))
+    const { documentText, vendorId, tenantId } = body as {
+      documentText?: string
+      vendorId?: string
+      tenantId?: string
+    }
 
-    // In production:
-    // 1. Fetch contract from database
-    // 2. Extract text from PDF/DOCX using embedding-service
-    // 3. Send to LLM for analysis
-    // 4. Store findings in database
-    // 5. Update contract with compliance score
+    // If documentText is provided, attempt LLM-based analysis
+    if (documentText) {
+      try {
+        // Fetch RAG context for contract review
+        const ragContext = await queryRAG(
+          'AVV Art. 28 DSGVO Auftragsverarbeitung Vertragsanforderungen',
+          3,
+          'bp_compliance_recht'
+        )
 
-    // For demo, return mock analysis results
+        // Build system prompt with RAG context
+        let systemPrompt = CONTRACT_REVIEW_SYSTEM_PROMPT
+        if (ragContext) {
+          systemPrompt += `\n\nRECHTSKONTEXT (als Referenz):\n${ragContext}`
+        }
+
+        // Call Ollama
+        const ollamaResponse = await fetch(`${OLLAMA_URL}/api/chat`, {
+          method: 'POST',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({
+            model: LLM_MODEL,
+            messages: [
+              { role: 'system', content: systemPrompt },
+              { role: 'user', content: `Analysiere den folgenden Vertrag auf DSGVO-Konformitaet:\n\n${documentText}` },
+            ],
+            stream: false,
+            options: { temperature: 0.1, num_predict: 16384 },
+            format: 'json',
+          }),
+          signal: AbortSignal.timeout(180000),
+        })
+
+        if (!ollamaResponse.ok) {
+          throw new Error(`LLM nicht erreichbar (Status ${ollamaResponse.status})`)
+        }
+
+        const result = await ollamaResponse.json()
+        const content = result.message?.content || ''
+        const llmResponse = JSON.parse(content)
+
+        // Transform LLM response to typed findings
+        const analysisResult = transformAnalysisResponse(llmResponse, {
+          contractId,
+          vendorId: vendorId || 'unknown',
+          tenantId: tenantId || 'default',
+          documentText,
+        })
+
+        return NextResponse.json({
+          success: true,
+          data: {
+            contractId,
+            findings: analysisResult.findings,
+            complianceScore: analysisResult.complianceScore,
+            reviewCompletedAt: new Date().toISOString(),
+            topRisks: analysisResult.topRisks,
+            requiredActions: analysisResult.requiredActions,
+            metadata: analysisResult.metadata,
+            parties: analysisResult.parties,
+            source: 'llm',
+          },
+          timestamp: new Date().toISOString(),
+        })
+      } catch (error) {
+        console.warn('LLM contract review failed, falling back to mock:', (error as Error).message)
+        // Fall through to mock findings
+      }
+    }
+
+    // Fallback: Mock analysis results
     const mockFindings: Finding[] = [
       {
         id: uuidv4(),
@@ -152,6 +228,7 @@ export async function POST(
           { de: 'Meldefrist auf 24-48h verkürzen', en: 'Reduce notification deadline to 24-48h' },
           { de: 'TIA für USA-Transfer durchführen', en: 'Conduct TIA for USA transfer' },
         ],
+        source: 'mock',
       },
       timestamp: new Date().toISOString(),
     })