enable filtering based on file name

app_stream_rag/app/chains.py

@@ -54,7 +54,12 @@ input_to_rag_chain = input_1 | input_2
 # HistoryInput and the LLM to build the rag_chain.
 rag_chain = (input_to_rag_chain | rag_prompt_formatted | llm).with_types(input_type=schemas.HistoryInput)
 
-# TODO:  Implement the filtered_rag_chain. It should be the 
+# Implement the filtered_rag_chain. It should be the 
 # same as the rag_chain but with hybrid_search = True.
-filtered_rag_chain = None
+input_2_filtered = {
+    'context': lambda x: format_context(data_indexer.search(x['new_question'], hybrid_search=True)),
+    'standalone_question': lambda x: x['new_question']
+}
+input_to_filtered_rag_chain = input_1 | input_2_filtered
+filtered_rag_chain = (input_to_filtered_rag_chain | rag_prompt_formatted | llm).with_types(input_type=schemas.HistoryInput)
 

app_stream_rag/app/main.py

@@ -14,7 +14,8 @@ from chains import (
     simple_chain,
     formatted_chain,
     history_chain,
-    rag_chain
+    rag_chain,
+    filtered_rag_chain
 )
 
 import models
@@ -148,6 +149,38 @@ async def rag_stream(request: Request, db: Session = Depends(get_db)):
         callbacks=[LogResponseCallback(user_request=user_request, db=db)]
         ))
 
+@app.post("/filtered_rag/stream")
+async def rag_stream(request: Request, db: Session = Depends(get_db)):  
+    # Receive request that had hit the endpoint
+    data = await request.json()
+    # Parse request into a user request
+    user_request = schemas.UserRequest(**data['input'])
+    username = user_request.username
+    question = user_request.question
+
+    # Pull the chat history of the user based on the user request
+    user_messages = crud.get_user_chat_history(db, username)
+    
+    # Use add_message & add the current question as part of the user history
+    message = schemas.MessageBase(
+        message=question,
+        type= "user",
+        timestamp=datetime.now()
+    )
+    
+    crud.add_message(db, message, username)
+
+    # create an instance of HistoryInput by using format_chat_history
+    user_chat_history = prompts.format_chat_history(user_messages)
+    history_input = schemas.HistoryInput(question=question, chat_history=user_chat_history)
+
+    # Use the history input within the rag chain
+    return EventSourceResponse(generate_stream(
+        history_input, 
+        filtered_rag_chain, 
+        callbacks=[LogResponseCallback(user_request=user_request, db=db)]
+        ))
+
 if __name__ == "__main__":
     import uvicorn
     uvicorn.run("main:app", host="localhost", reload=True,  port=8000)

app_stream_rag/test.py

@@ -3,7 +3,7 @@ from langserve import RemoteRunnable
 # If we put /simple/stream, it complains; because chain.stream will hit /simple/stream endpoint
 url = "https://simafarazi-backend-c.hf.space/rag"
 chain = RemoteRunnable(url) #Client for iteracting with LangChain runnables that are hosted as LangServe endpoints
-stream = chain.stream(input={"question":"Where do you recommend me for a 7 days hike in europe during April?", 
+stream = chain.stream(input={"question":"What does the following code do? input_1 = RunnablePassthrough.assign(new_question=standalone_chain)", 
                              "username": "Sima"}) # .stream() and .invoke() are standard methods to interact with hosted runnables