Fixes

App/main.py

@@ -449,8 +449,23 @@ async def query_rag_endpoint(payload: QueryModel):
         else:
             user_lan = language
         question_tokens = count_tokens(payload.question)
-        print(f"Question tokens: {question_tokens}")
-        retrieved_docs = app.state.retriever.get_relevant_documents(payload.question)
+
+
+
+
+        # In your query_rag_endpoint, before rag_chain.invoke():
+        try:
+            retrieved_docs = app.state.retriever.get_relevant_documents(payload.question)
+        except Exception as e:
+            # Rebuild retriever and try again
+            app.state.retriever = get_existing_retriever()
+            if not app.state.retriever:
+                raise HTTPException(400, "Vector store corrupted. Please rebuild.")
+            retrieved_docs = app.state.retriever.get_relevant_documents(payload.question)
+
+
+
+        # retrieved_docs = app.state.retriever.get_relevant_documents(payload.question)
         context_text = "\n\n".join([doc.page_content for doc in retrieved_docs])
         context_tokens = count_tokens(context_text)
         chat_history = await chatbot_history_collection.find_one({"userId": payload.user_id})

App/utils.py

@@ -19,6 +19,17 @@ import tiktoken
 from googleapiclient.discovery import build
 import numpy as np
 
+
+
+os.environ['TRANSFORMERS_CACHE'] = '/tmp/transformers_cache'
+os.environ['HF_HOME'] = '/tmp/huggingface_home'
+os.environ['HUGGINGFACE_HUB_CACHE'] = '/tmp/huggingface_cache'
+
+# Create the directories
+os.makedirs('/tmp/transformers_cache', exist_ok=True)
+os.makedirs('/tmp/huggingface_home', exist_ok=True)
+os.makedirs('/tmp/huggingface_cache', exist_ok=True)
+
 # MongoDB connection
 MONGODB_URI = os.getenv(
     "MONGODB_URI",