Update app.py

app.py

@@ -98,6 +98,56 @@ def query_document(question):
     tts.save(temp_audio_path)
     return response, temp_audio_path
 
+def query_document(question):
+    if vector_store is None:
+        return "❌ No documents processed.", None
+
+    # ✅ Fetch stored documents  
+    stored_docs = vector_store.get()["documents"]  
+
+    # ✅ Calculate total word count safely  
+    total_words = sum(len(doc.split()) if isinstance(doc, str) else len(doc.page_content.split()) for doc in stored_docs)  
+
+    # ✅ Dynamically adjust k based on document size  
+    if total_words < 500:  
+        k_value = 3  
+    elif total_words < 2000:  
+        k_value = 5  
+    else:  
+        k_value = 10  
+
+    retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": k_value})  
+
+    # ✅ Improved prompt for detailed response  
+    detailed_prompt = f"""  
+    Provide a **detailed and structured answer** to the following question.  
+    - Use relevant **examples, key points, and explanations**.  
+    - If applicable, provide **step-by-step analysis** or comparisons.  
+    - Ensure **clarity and completeness**.  
+    **Question:** {question}  
+    """  
+
+    # ✅ Dynamically select model based on document size  
+    if total_words < 1000:  
+        model_name = "gemini-2.0-pro-exp-02-05"  # More detailed responses for small files  
+    else:  
+        model_name = "gemini-2.0-flash"  # Faster processing for large documents  
+
+    logging.info(f"🧠 Using Model: {model_name} for processing")  
+
+    model = ChatGoogleGenerativeAI(model=model_name, google_api_key=GOOGLE_API_KEY)  
+    qa_chain = RetrievalQA.from_chain_type(llm=model, retriever=retriever)  
+    response = qa_chain.invoke({"query": detailed_prompt})["result"]  
+
+    # ✅ Convert response to speech  
+    tts = gTTS(text=response, lang="en")  
+    temp_audio_path = os.path.join(temp_dir, "response.mp3")  
+    tts.save(temp_audio_path)  
+    temp_file_map["response.mp3"] = time.time()  
+
+    return response, temp_audio_path 
+
+
 # ✅ Gradio UI
 with gr.Blocks() as demo:
     gr.Markdown("# 📄 AI-Powered Multi-Document Chatbot with Voice Output")