Update app.py

app.py

@@ -1,12 +1,12 @@
 import gradio as gr
 import re
 from datetime import datetime
-from transformers import pipeline, Conversation
+from transformers import pipeline
 
-# Initialize the conversational AI
+# Initialize the conversational AI with a model that works in Hugging Face Spaces
 chatbot = pipeline(
     "conversational", 
-    model="facebook/blenderbot-400M-distill",  # More conversational model
+    model="facebook/blenderbot-400M-distill",
     max_length=200
 )
 
@@ -33,8 +33,11 @@ def initialize_state():
         "stage": ChatState.NORMAL,
         "return_product": None,
         "receipt": None,
-        "conversation": Conversation(),
-        "error_count": 0
+        "name": None,
+        "purchase_date": None,
+        "email": None,
+        "error_count": 0,
+        "conversation_history": []  # Store conversation as tuples
     }
 
 # Helper functions
@@ -42,14 +45,25 @@ def process_return(product):
     # Simulate return processing
     return f"✅ Return processed for your {product}!\nConfirmation sent to your email."
 
-def handle_return_flow(user_input, state, conversation):
+def is_return_request(text):
+    return bool(re.search(r'\b(return|refund|exchange|send back)\b', text.lower()))
+
+def generate_chat_prompt(history, new_input):
+    """Build a conversational prompt for the AI model"""
+    prompt = ""
+    for user, bot in history:
+        prompt += f"User: {user}\nAssistant: {bot}\n"
+    prompt += f"User: {new_input}\nAssistant:"
+    return prompt
+
+def handle_return_flow(user_input, state):
     """Natural conversation handling for return process"""
     response = ""
     new_stage = state["stage"]
     
     # Detect return request naturally
     if state["stage"] == ChatState.NORMAL:
-        if re.search(r'\b(return|refund|exchange|send back)\b', user_input.lower()):
+        if is_return_request(user_input):
             state["stage"] = ChatState.RETURN_DETECTED
             response = "I'd be happy to help with your return! Could you tell me which product you'd like to return?"
     
@@ -114,7 +128,6 @@ def handle_return_flow(user_input, state, conversation):
         response += "\n\nIs there anything else I can help you with today?"
         # Reset conversation but keep context
         state = initialize_state()
-        state["conversation"] = conversation  # Maintain conversation history
     
     # Disengagement
     elif state["stage"] == ChatState.DISENGAGED:
@@ -124,34 +137,22 @@ def handle_return_flow(user_input, state, conversation):
 
 def chat_fn(user_input, chat_history, state):
     """Main chat function with natural conversation handling"""
-    # Initialize conversation if needed
-    if not state["conversation"].past_user_inputs:
-        state["conversation"].add_user_input("Hello!")
-        state["conversation"] = chatbot(state["conversation"])
-    
-    # Add user message to conversation
-    state["conversation"].add_user_input(user_input)
+    # First check if we're in a return flow
+    return_response, state = handle_return_flow(user_input, state)
     
-    # Handle return flow naturally within conversation
-    return_response, state = handle_return_flow(user_input, state, state["conversation"])
-    
-    # If we have a specific response from return flow, use it
     if return_response:
+        # We have a response from the return flow
         response = return_response
-        state["conversation"].append_response(response)
     else:
-        # Generate conversational response
-        state["conversation"] = chatbot(state["conversation"])
-        response = state["conversation"].generated_responses[-1]
+        # Generate conversational response using AI
+        prompt = generate_chat_prompt(state["conversation_history"], user_input)
+        result = chatbot(prompt, max_length=200)
+        response = result[0]['generated_text'].split("Assistant:")[-1].strip()
     
-    # Update chat history
+    # Update conversation history
+    state["conversation_history"].append((user_input, response))
     chat_history.append((user_input, response))
     
-    # Reset state if we completed return process
-    if state["stage"] == ChatState.RETURN_PROCESSING:
-        state = initialize_state()
-        state["conversation"].append_response(response)  # Maintain context
-    
     return "", chat_history, state
 
 # Gradio interface with natural conversation design