ai-service-admin/src/api/rag.ts

@@ -1,4 +1,5 @@
 import request from '@/utils/request'
+import { useTenantStore } from '@/stores/tenant'
 
 export interface AIResponse {
   content: string
@@ -73,6 +74,8 @@ export function createSSEConnection(
   const baseUrl = import.meta.env.VITE_APP_BASE_API || '/api'
   const fullUrl = `${baseUrl}${url}`
   
+  const tenantStore = useTenantStore()
+  
   const controller = new AbortController()
   
   fetch(fullUrl, {
@@ -80,6 +83,7 @@ export function createSSEConnection(
     headers: {
       'Content-Type': 'application/json',
       'Accept': 'text/event-stream',
+      'X-Tenant-Id': tenantStore.currentTenantId || '',
     },
     body: JSON.stringify(body),
     signal: controller.signal

ai-service/app/core/config.py

@@ -44,9 +44,16 @@ class Settings(BaseSettings):
     ollama_embedding_model: str = "nomic-embed-text"
 
     rag_top_k: int = 5
-    rag_score_threshold: float = 0.3
+    rag_score_threshold: float = 0.01
     rag_min_hits: int = 1
     rag_max_evidence_tokens: int = 2000
+    
+    rag_two_stage_enabled: bool = True
+    rag_two_stage_expand_factor: int = 10
+    rag_hybrid_enabled: bool = True
+    rag_rrf_k: int = 60
+    rag_vector_weight: float = 0.7
+    rag_bm25_weight: float = 0.3
 
     confidence_low_threshold: float = 0.5
     confidence_high_threshold: float = 0.8

ai-service/app/services/retrieval/vector_retriever.py

@@ -61,20 +61,31 @@ class VectorRetriever(BaseRetriever):
             RetrievalResult with filtered hits.
         """
         logger.info(
-            f"[AC-AISVC-16] Starting vector retrieval for tenant={ctx.tenant_id}, query={ctx.query[:50]}..."
+            f"[AC-AISVC-16] Starting vector retrieval for tenant={ctx.tenant_id}, "
+            f"query={ctx.query[:50]}..."
+        )
+        logger.info(
+            f"[AC-AISVC-16] Retrieval config: top_k={self._top_k}, "
+            f"score_threshold={self._score_threshold}, min_hits={self._min_hits}"
         )
 
         try:
             client = await self._get_client()
+            logger.info(f"[AC-AISVC-16] Got Qdrant client: {type(client).__name__}")
 
+            logger.info("[AC-AISVC-16] Generating embedding for query...")
             query_vector = await self._get_embedding(ctx.query)
+            logger.info(f"[AC-AISVC-16] Embedding generated: dim={len(query_vector)}")
 
+            logger.info(f"[AC-AISVC-16] Searching in tenant collection: tenant_id={ctx.tenant_id}")
             hits = await client.search(
                 tenant_id=ctx.tenant_id,
                 query_vector=query_vector,
                 limit=self._top_k,
                 score_threshold=self._score_threshold,
             )
+            
+            logger.info(f"[AC-AISVC-16] Search returned {len(hits)} raw hits")
 
             retrieval_hits = [
                 RetrievalHit(
@@ -104,6 +115,12 @@ class VectorRetriever(BaseRetriever):
                 f"[AC-AISVC-17] Retrieval complete: {len(retrieval_hits)} hits, "
                 f"insufficient={is_insufficient}, max_score={diagnostics['max_score']:.3f}"
             )
+            
+            if len(retrieval_hits) == 0:
+                logger.warning(
+                    f"[AC-AISVC-17] No hits found! tenant={ctx.tenant_id}, "
+                    f"query={ctx.query[:50]}..., raw_hits={len(hits)}, threshold={self._score_threshold}"
+                )
 
             return RetrievalResult(
                 hits=retrieval_hits,
@@ -111,7 +128,7 @@ class VectorRetriever(BaseRetriever):
             )
 
         except Exception as e:
-            logger.error(f"[AC-AISVC-16] Retrieval error: {e}")
+            logger.error(f"[AC-AISVC-16] Retrieval error: {e}", exc_info=True)
             return RetrievalResult(
                 hits=[],
                 diagnostics={"error": str(e), "is_insufficient": True},