Update app.py

app.py

@@ -237,7 +237,7 @@ class ProductionRAGSystem:
             status_text.empty()
             return False
     
-    def search(self, query, n_results=3):
+    def search(self, query, n_results=5):  # Increased from 3 to 5
         """Search for relevant chunks"""
         if not self.model or not self.collection:
             return None
@@ -266,6 +266,12 @@ class ProductionRAGSystem:
                     'similarity': similarity
                 })
             
+            # Debug: Show search results for troubleshooting
+            print(f"Search for '{query}' found {len(search_results)} results")
+            for i, result in enumerate(search_results[:3]):
+                print(f"  {i+1}. Similarity: {result['similarity']:.3f} | Source: {result['metadata']['source_file']}")
+                print(f"     Content preview: {result['content'][:100]}...")
+            
             return search_results
         except Exception as e:
             st.error(f"Search error: {e}")
@@ -746,10 +752,16 @@ if prompt := st.chat_input("Ask questions about your documents..."):
     with st.chat_message("assistant"):
         if rag_system and rag_system.model and rag_system.get_collection_count() > 0:
             # Search documents first
-            search_results = rag_system.search(prompt, n_results=3)
+            search_results = rag_system.search(prompt, n_results=5)
+            
+            # Debug output for troubleshooting
+            if search_results:
+                st.info(f"🔍 Found {len(search_results)} potential matches. Best similarity: {search_results[0]['similarity']:.3f}")
+            else:
+                st.warning("🔍 No search results returned from vector database")
             
-            # Check if we found relevant documents (lower threshold for better coverage)
-            if search_results and search_results[0]['similarity'] > 0.005:  # Even lower threshold
+            # Check if we found relevant documents (very low threshold)
+            if search_results and search_results[0]['similarity'] > 0.001:  # Ultra-low threshold
                 # Generate document-based answer
                 result = rag_system.generate_answer(
                     prompt, 
@@ -775,14 +787,15 @@ if prompt := st.chat_input("Ask questions about your documents..."):
                     if use_ai_enhancement and not result['has_both']:
                         st.info("💡 AI enhancement failed - showing extracted answer from documents")
                 
-                # Show RAG info
+                # Show RAG info with more details
                 if show_sources and result['sources']:
                     confidence_text = f"{result['confidence']*100:.1f}%" if show_confidence else ""
                     st.markdown(f"""
                     <div class="rag-attribution">
                         <strong>📁 Sources:</strong> {', '.join(result['sources'])}<br>
                         <strong>🎯 Confidence:</strong> {confidence_text}<br>
-                        <strong>📊 Found:</strong> {len(search_results)} relevant sections
+                        <strong>📊 Found:</strong> {len(search_results)} relevant sections<br>
+                        <strong>🔍 Best Match:</strong> {search_results[0]['similarity']:.3f} similarity
                     </div>
                     """, unsafe_allow_html=True)
                 
@@ -799,8 +812,12 @@ if prompt := st.chat_input("Ask questions about your documents..."):
                 }
                 
             else:
-                # No relevant documents found - use general AI
-                st.info("No relevant documents found. Using general AI mode...")
+                # No relevant documents found - show debug info
+                if search_results:
+                    st.warning(f"📄 Found documents but similarity too low (best: {search_results[0]['similarity']:.3f}). Using general AI...")
+                else:
+                    st.warning("📄 No documents found in search. Using general AI...")
+                    
                 general_response = get_general_ai_response(prompt, unlimited_tokens=unlimited_tokens)
                 st.markdown(f"💬 **General AI:** {general_response}")