Update app.py

app.py

@@ -80,7 +80,7 @@ async def chat(request: ChatRequest):
         if request.model_choice == "google":
             client = genai.Client(api_key=google_api_key)
 
-            # summarize chat history
+            # ------------ summarize chat history ------------
             summary_thresh = 10
             if len(request.chat_history) > summary_thresh:
                 summarize_prompt = f"""Please summarize the following chat history concisely, focusing on the key points and main topics discussed. Avoid 
@@ -98,12 +98,42 @@ async def chat(request: ChatRequest):
                 request.chat_history.insert(1, 
                     {"role": "user", 
                      "parts": [{"text": f"Here is a summary of this conversation so far: {summary_response.text}"}]})
-            
 
-            # Retrieve relevant text
-            docs, filtered_docs_and_scores = retrieve(request.message, vectorstore, top_k=8)
+            # ------------ rephrase user question ------------
+            rephrase_prompt = f"""You are an AI that enhances user questions for better retrieval in a Retrieval-Augmented Generation (RAG) system. Given a vague user question, rewrite it as a rich, context-filled question that provides specificity, improves retrieval accuracy, and ensures detailed responses.
+
+                    Expand on key concepts related to the original question.
+                    
+                    Add relevant context, such as location, conditions, behavior patterns, or influencing factors.
+                    
+                    Keep the core intent of the user's question but refine it for depth and clarity.
+                    
+                    Example Inputs & Outputs:
+                    
+                    User Input: "Where do bucks bed?"
+                    Enhanced Question: "What types of terrain and cover do bucks prefer for bedding, and how do factors like pressure, wind, and thermals influence their bedding locations?"
+                    
+                    User Input: "How do I get success early season?"
+                    Enhanced Question: "What key factors contribute to early-season hunting success, including buck bedding habits, food sources, and weather patterns, and how can hunters adjust their strategy to capitalize on predictable movement?"
+                    
+                    Now, transform the following user question into a rich, context-filled question:
+                    {request.user_message}"""
+            
+            rephrase_response = client.models.generate_content(
+                model="gemini-2.0-flash",
+                contents=rephrase_prompt,
+                config=GenerateContentConfig(
+                    system_instruction=["You are a helpful assistant who is an expert at summarization."],
+                    max_output_tokens=250,
+                    temperature=0.5
+                ),
+            )
+            
+            #  ------------ Retrieve relevant text ------------
+            docs, filtered_docs_and_scores = retrieve(rephrase_response, vectorstore, top_k=8)
             docs = "\n\n".join(docs)
 
+            # ------------ Retrievel Augmented Generation ------------
             rag_prompt = f"""Use the following information to answer the user's query. You do not have to use all the information, just the pieces that directly 
             help answer the query most accurately. Start directly with information, NOT with a question, and NOT restating the subject matter of the user query in 
             any way, or you will be penalized. Respond in a conversational manner. 
@@ -126,7 +156,7 @@ async def chat(request: ChatRequest):
 
             Using the information above, answer the user's query as accurately as possible in the tone and style of the Good Response:
 
-            User Query: {request.message}
+            User Query: {rephrase_response}
             """
             
             # remove the unfformatted user message     
@@ -149,7 +179,7 @@ async def chat(request: ChatRequest):
             del request.chat_history[-1]
             request.chat_history.append({"role": "user", "parts": [{"text": request.message}]})
 
-            return {"response": response.text, "dataset_str": text_string, "docs": docs, "filtered_docs_and_scores": filtered_docs_and_scores, "history": request.chat_history, "RAG_prompt": rag_prompt, "chunks": chunks}
+            return {"response": response.text, "dataset_str": text_string, "rephrase_response", rephrase_response, "docs": docs, "filtered_docs_and_scores": filtered_docs_and_scores, "history": request.chat_history, "RAG_prompt": rag_prompt, "chunks": chunks}
 
         if request.model_choice == "HF":
             if hf_token: