Update app.py

app.py

@@ -35,33 +35,29 @@ class ChatRequest(BaseModel):
 class SummaryRequest(BaseModel):
     chat_history: list  # List of messages
 
+
+# Load Local LLM (Mistral or Llama)
+model_name = "mistralai/Mistral-7B-Instruct-v0.3"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name)
+
+def generate_local_emotional_response(user_input, questions):
+    """Generate emotional responses locally using LLaMA/Mistral."""
+    prompt = f"User: {user_input}\n\nBased on this, respond in an empathetic way before asking each question:\n1. {questions[0]}\n2. {questions[1]}\n3. {questions[2]}"
+    inputs = tokenizer(prompt, return_tensors="pt")
+    output = model.generate(**inputs, max_length=200)
+    return tokenizer.decode(output[0], skip_special_tokens=True).split("\n")
+
 @app.post("/get_questions")
 def get_recommended_questions(request: ChatRequest):
-    """Retrieve the most relevant diagnostic questions with a conversational response."""
-    
-    # Step 1: Encode the input message for FAISS search
     input_embedding = embedding_model.encode([request.message], convert_to_numpy=True)
     distances, indices = question_index.search(input_embedding, 3)
-
-    # Step 2: Retrieve the top 3 relevant questions
     retrieved_questions = [questions_df["Questions"].iloc[i] for i in indices[0]]
 
-    # Step 3: Define Dynamic Prompt Variations
-    empathetic_phrases = [
-        "I hear you, and I appreciate you sharing this. Let me ask:",
-        "That sounds challenging. Could you help me understand better by answering this:",
-        "I understand what you're going through. Here's something to think about:",
-        "Thank you for opening up. Let’s explore this further:",
-        "Your feelings are completely valid. Here's a question to help us understand more:"
-    ]
-
-    # Step 4: Generate Dynamic Responses
-    wrapped_responses = [
-        f"{random.choice(empathetic_phrases)} *{q}*"
-        for q in retrieved_questions
-    ]
-
-    return {"questions": wrapped_responses}
+    # Generate dynamic emotional responses locally
+    enhanced_responses = generate_local_emotional_response(request.message, retrieved_questions)
+
+    return {"questions": enhanced_responses}
 
 @app.post("/summarize_chat")
 def summarize_chat(request: SummaryRequest):