fixed bugs

app/services/rag.py

@@ -11,11 +11,11 @@ RAG_SYSTEM_PROMPT = """You are a helpful AI assistant for {company_name}.
 Your role is to answer questions based on the provided context from company documents.
 
 IMPORTANT RULES:
-1. Only answer based on the provided context
-2. If information is not in the context, say "I don't have information about that in my knowledge base"
+1. Answer based on the provided context below
+2. If the context does not contain enough information, say so, but also try to be helpful with what IS available
 3. Be concise and helpful
 4. Always maintain a professional, friendly tone
-5. If asked about topics outside the context, politely redirect to relevant topics
+5. If asked about topics completely outside the context, politely redirect to relevant topics
 
 {custom_instructions}
 
@@ -47,8 +47,9 @@ class RAGEngine:
         # Step 1: Embed the query
         try:
             query_embedding = self.embedding_svc.embed_text(query)
+            logger.info(f"[RAG] Query embedded successfully. Vector length: {len(query_embedding)}")
         except Exception as e:
-            logger.error(f"Embedding error: {e}")
+            logger.error(f"[RAG] Embedding error: {e}", exc_info=True)
             return {
                 "response": "I'm having trouble processing your request. Please try again.",
                 "sources": [],
@@ -57,13 +58,22 @@ class RAGEngine:
             }
 
         # Step 2: Retrieve relevant chunks
+        # FIX: Lowered score_threshold from 0.3 to 0.1 to avoid filtering out
+        # all results. With cosine similarity, 0.3 can be too aggressive for
+        # many document types and query patterns.
         retrieved = self.vector_svc.search(
             collection_name=collection_name,
             query_vector=query_embedding,
             limit=5,
-            score_threshold=0.3,
+            score_threshold=0.1,  # FIX: was 0.3, now 0.1 to avoid over-filtering
         )
 
+        logger.info(f"[RAG] Retrieved {len(retrieved)} chunks from collection '{collection_name}'")
+        for i, item in enumerate(retrieved):
+            score = item.get("score", 0)
+            text_preview = item.get("payload", {}).get("text", "")[:80]
+            logger.info(f"[RAG]   Chunk {i+1}: score={score:.4f}, preview='{text_preview}...'")
+
         # Step 3: Build sources
         sources = []
         context_parts = []
@@ -84,7 +94,12 @@ class RAGEngine:
                     )
                 )
 
-        context = "\n\n---\n\n".join(context_parts) if context_parts else "No relevant information found."
+        if context_parts:
+            context = "\n\n---\n\n".join(context_parts)
+            logger.info(f"[RAG] Built context from {len(context_parts)} chunks ({len(context)} chars)")
+        else:
+            context = "No relevant information found in the knowledge base."
+            logger.warning(f"[RAG] No context found for query: '{query}' in collection '{collection_name}'")
 
         # Step 4: Build messages
         system_prompt = RAG_SYSTEM_PROMPT.format(
@@ -95,13 +110,18 @@ class RAGEngine:
 
         messages = [{"role": "system", "content": system_prompt}]
 
-        # Add conversation history (last 10 messages)
-        for msg in conversation_history[-10:]:
+        # FIX: Conversation history must be in CHRONOLOGICAL order (oldest first).
+        # The history should already come sorted ascending from the chat router.
+        # We take the last 10 messages for context window management.
+        history_to_use = conversation_history[-10:] if conversation_history else []
+        for msg in history_to_use:
             messages.append({"role": msg["role"], "content": msg["content"]})
 
         # Add current query
         messages.append({"role": "user", "content": query})
 
+        logger.info(f"[RAG] Sending {len(messages)} messages to LLM (model: {chatbot_config.get('model')})")
+
         # Step 5: Generate response
         model = chatbot_config.get("model", "accounts/fireworks/models/kimi-k2-instruct-0905")
         try:
@@ -111,6 +131,7 @@ class RAGEngine:
                 max_tokens=chatbot_config.get("max_tokens", 1000),
                 temperature=chatbot_config.get("temperature", 0.7),
             )
+            logger.info(f"[RAG] LLM response generated. Tokens used: {result.get('tokens_used', 0)}")
             return {
                 "response": result["content"],
                 "sources": sources,
@@ -118,7 +139,7 @@ class RAGEngine:
                 "model": result.get("model", model),
             }
         except Exception as e:
-            logger.error(f"LLM generation error: {e}")
+            logger.error(f"[RAG] LLM generation error: {e}", exc_info=True)
             return {
                 "response": "I'm having trouble generating a response. Please try again later.",
                 "sources": sources,
@@ -127,4 +148,4 @@ class RAGEngine:
             }
 
 
-rag_engine = RAGEngine()
+rag_engine = RAGEngine()