made changes to add_document fxn

app/main_api.py

@@ -778,20 +778,68 @@ class DeadlockFreeRAGPipeline:
         logger.info(f"🚀 Embedding {len(texts)} chunks via Kaggle...")
         embeddings = await self.kaggle_client.generate_embeddings(texts)
         
+        # --- ADD THIS DEBUGGING SECTION ---
+        logger.info("--- HF DEBUG ---")
+        logger.info(f"Type of embeddings received: {type(embeddings)}")
+        if isinstance(embeddings, list) and len(embeddings) > 0:
+            logger.info(f"Number of embeddings: {len(embeddings)}")
+            logger.info(f"Type of first item: {type(embeddings[0])}")
+            if isinstance(embeddings[0], list):
+                logger.info(f"Dimension of first embedding: {len(embeddings[0])}")
+                logger.info(f"First few values: {embeddings[0][:5] if len(embeddings[0]) > 5 else embeddings[0]}")
+        else:
+            logger.info("Embeddings variable is either not a list or is empty!")
+        logger.info("--- END HF DEBUG ---")
+        # ---------------------------------
+        
         if not embeddings or len(embeddings) != len(texts):
             logger.error("Embedding failed or returned mismatched count.")
+            logger.error(f"Expected {len(texts)} embeddings, got {len(embeddings) if embeddings else 0}")
+            return
+        
+        # Validate embeddings format
+        if not isinstance(embeddings, list):
+            logger.error(f"Embeddings must be a list, got {type(embeddings)}")
             return
         
+        # Check if all embeddings are lists of floats
+        for i, emb in enumerate(embeddings[:3]):  # Check first 3
+            if not isinstance(emb, list):
+                logger.error(f"Embedding {i} is not a list: {type(emb)}")
+                return
+            if not all(isinstance(x, (int, float)) for x in emb[:5]):  # Check first 5 values
+                logger.error(f"Embedding {i} contains non-numeric values")
+                return
+        
         # GEMINI'S FIX: Step 3 - Add to Chroma with pre-calculated embeddings
         # This completely avoids the deadlock!
-        self.vectorstore.add_texts(
-            texts=texts,
-            metadatas=[chunk['metadata'] for chunk in quality_chunks[:100]],
-            embeddings=embeddings  # <-- THE CRITICAL FIX - MAKE SURE THIS IS HERE
-        )
-        
-        logger.info(f"✅ Added {len(texts)} documents with embeddings to vector store (DEADLOCK-FREE)")
-         
+        try:
+            logger.info("🔄 Adding texts to Chroma with embeddings...")
+            self.vectorstore.add_texts(
+                texts=texts,
+                metadatas=[chunk['metadata'] for chunk in quality_chunks[:100]],
+                embeddings=embeddings  # <-- THE CRITICAL FIX - MAKE SURE THIS IS HERE
+            )
+            logger.info(f"✅ Added {len(texts)} documents with embeddings to vector store (DEADLOCK-FREE)")
+            
+        except Exception as e:
+            logger.error(f"❌ ChromaDB add_texts failed: {e}")
+            logger.error(f"❌ Error type: {type(e)}")
+            
+            # Try adding without embeddings as fallback (this will show if it's an embedding format issue)
+            try:
+                logger.info("🔄 Trying to add texts without embeddings (fallback)...")
+                self.vectorstore.add_texts(
+                    texts=texts,
+                    metadatas=[chunk['metadata'] for chunk in quality_chunks[:100]]
+                    # No embeddings parameter - let Chroma handle it
+                )
+                logger.info("✅ Fallback successful - issue is with embedding format")
+            except Exception as fallback_error:
+                logger.error(f"❌ Even fallback failed: {fallback_error}")
+            
+            raise e  # Re-raise original error
+
     async def answer_question(self, question: str) -> str:
         """GEMINI'S FIX: Direct query embedding - no deadlock"""
         # Security check