Update app.py

app.py

@@ -15,12 +15,12 @@ import re
 st.set_page_config(page_title="RAG Customer Support Chatbot", layout="wide")
 
 # --- Default Configurations & File Paths ---
-DEFAULT_TWILIO_ACCOUNT_SID_FALLBACK = "" # Fallback if secret "TWILIO_SID" is not found
-DEFAULT_TWILIO_AUTH_TOKEN_FALLBACK = "" # Fallback if secret "TWILIO_TOKEN" is not found
-DEFAULT_GROQ_API_KEY_FALLBACK = "" # Fallback if secret "GROQ_API_KEY" is not found
+DEFAULT_TWILIO_ACCOUNT_SID_FALLBACK = "" 
+DEFAULT_TWILIO_AUTH_TOKEN_FALLBACK = "" 
+DEFAULT_GROQ_API_KEY_FALLBACK = "" 
 
 DEFAULT_TWILIO_CONVERSATION_SERVICE_SID = ""
-DEFAULT_TWILIO_BOT_WHATSAPP_IDENTITY = st.secrets.get("TWILIO_PHONE_NUMBER", "whatsapp:+14155238886") # Twilio Sandbox default
+DEFAULT_TWILIO_BOT_WHATSAPP_IDENTITY = st.secrets.get("TWILIO_PHONE_NUMBER", "whatsapp:+14155238886") 
 DEFAULT_EMBEDDING_MODEL_NAME = "sentence-transformers/all-MiniLM-L6-v2"
 DEFAULT_POLLING_INTERVAL_S = 30
 DOCS_FOLDER = "docs/"
@@ -30,19 +30,16 @@ POLICY_PDF_FILE = os.path.join(DOCS_FOLDER, "ProductReturnPolicy.pdf")
 FAQ_PDF_FILE = os.path.join(DOCS_FOLDER, "FAQ.pdf")
 
 # --- Application Secrets Configuration ---
-# These are the primary keys fetched from st.secrets as per user request
 APP_TWILIO_ACCOUNT_SID = st.secrets.get("TWILIO_ACCOUNT_SID")
 APP_TWILIO_AUTH_TOKEN = st.secrets.get("TWILIO_AUTH_TOKEN")
 APP_GROQ_API_KEY = st.secrets.get("GROQ_API_KEY")
 
-# Other secrets with fallback to defaults/sidebar input (if secrets not found)
 APP_TWILIO_CONVERSATION_SERVICE_SID_SECRET = st.secrets.get("TWILIO_CONVERSATION_SERVICE_SID")
 APP_TWILIO_BOT_WHATSAPP_IDENTITY_SECRET = st.secrets.get("TWILIO_BOT_WHATSAPP_IDENTITY")
 
 
 # --- RAG Processing Utilities ---
 def load_json_data(file_path):
-    """Loads data from a JSON file."""
     try:
         with open(file_path, 'r', encoding='utf-8') as f:
             data = json.load(f)
@@ -58,7 +55,6 @@ def load_json_data(file_path):
         return None
 
 def load_pdf_data(file_path):
-    """Extracts text from a PDF file, page by page."""
     try:
         with open(file_path, 'rb') as f:
             reader = PyPDF2.PdfReader(f)
@@ -75,7 +71,6 @@ def load_pdf_data(file_path):
         return []
 
 def chunk_text(text_pages, chunk_size=1000, chunk_overlap=200):
-    """Chunks text from PDF pages into smaller, overlapping pieces."""
     full_text = "\n".join(text_pages)
     if not full_text.strip():
         return []
@@ -93,7 +88,6 @@ def chunk_text(text_pages, chunk_size=1000, chunk_overlap=200):
 
 @st.cache_resource(show_spinner="Initializing embedding model...")
 def initialize_embedding_model(model_name=DEFAULT_EMBEDDING_MODEL_NAME):
-    """Initializes and returns a SentenceTransformer model."""
     try:
         model = SentenceTransformer(model_name)
         return model
@@ -103,7 +97,6 @@ def initialize_embedding_model(model_name=DEFAULT_EMBEDDING_MODEL_NAME):
 
 @st.cache_resource(show_spinner="Building FAISS index for PDF documents...")
 def create_faiss_index(_text_chunks, _embedding_model):
-    """Creates a FAISS index from text chunks and an embedding model."""
     if not _text_chunks or _embedding_model is None:
         st.warning("Cannot create FAISS index: No text chunks or embedding model available.")
         return None, []
@@ -127,7 +120,6 @@ def create_faiss_index(_text_chunks, _embedding_model):
         return None, []
 
 def search_faiss_index(index, query_text, embedding_model, indexed_chunks, k=3):
-    """Searches the FAISS index and returns top_k relevant chunk texts."""
     if index is None or embedding_model is None or not query_text:
         return []
     try:
@@ -146,7 +138,6 @@ def search_faiss_index(index, query_text, embedding_model, indexed_chunks, k=3):
         return []
 
 def get_order_details(order_id, customer_orders_data):
-    """Retrieves order details for a given order_id."""
     if not customer_orders_data:
         return "Customer order data is not loaded."
     for order in customer_orders_data:
@@ -155,7 +146,6 @@ def get_order_details(order_id, customer_orders_data):
     return f"No order found with ID: {order_id}."
 
 def get_product_info(query, products_data):
-    """Retrieves product information based on a query."""
     if not products_data:
         st.warning("Product data is not loaded or is empty in get_product_info.")
         return "Product data is not loaded."
@@ -195,11 +185,6 @@ def get_product_info(query, products_data):
 def generate_response_groq(_groq_client, query, context, model="llama3-8b-8192",
                            intent=None, customer_name=None, item_name=None,
                            shipping_address=None, delivery_date=None, order_id=None, order_status=None):
-    """
-    Generates a response using GROQ LLaMA3 API.
-    If intent is ORDER_STATUS and customer_name, order_id, order_status are provided,
-    it crafts a more personalized prompt.
-    """
     if not _groq_client:
         return "GROQ client not initialized. Please check API key."
     if not query:
@@ -209,34 +194,58 @@ def generate_response_groq(_groq_client, query, context, model="llama3-8b-8192",
     user_prompt = ""
 
     if intent == "ORDER_STATUS" and order_id and customer_name and order_status:
-        system_message = f"You are a friendly and helpful customer support assistant. Your primary goal is to provide a personalized update about order {order_id} for {customer_name}."
+        system_message = (
+            f"You are an exceptionally friendly and helpful customer support assistant. "
+            f"Your current task is to provide a single, complete, and human-like sentence as a response to {customer_name} "
+            f"about their order {order_id}. You MUST incorporate all relevant order details provided into this single sentence."
+        )
         
-        item_description = item_name if item_name else "your item(s)" # Default if item_name is not found
+        item_description = item_name if item_name else "the ordered item(s)"
         
-        # Base of the user prompt, asking the LLM to formulate the response
-        user_prompt = f"The user, {customer_name}, asked the following question: '{query}'.\n"
-        user_prompt += f"You have the following details for their order {order_id}:\n"
-        user_prompt += f"- Item(s): {item_description}\n"
-        user_prompt += f"- Current Status: '{order_status}'\n"
+        # Construct the core information string that the LLM needs to build upon
+        core_info_parts = [
+            f"your order {order_id}",
+            f"for {item_description}",
+            f"has a status of '{order_status}'"
+        ]
 
         if order_status.lower() == "delivered":
-            user_prompt += f"- Shipping Address: {shipping_address if shipping_address else 'their address'}\n"
-            user_prompt += f"- Delivered On: {delivery_date if delivery_date else 'the specified date'}\n"
-            user_prompt += f"\nPlease formulate a human-like response to {customer_name}. For example: "
-            user_prompt += f"'Hi {customer_name}, regarding your order {order_id} for {item_description}, it has been \"{order_status}\". "
-            if shipping_address and delivery_date:
-                 user_prompt += f"It was delivered to {shipping_address} on {delivery_date}.'"
-            elif shipping_address:
-                 user_prompt += f"It was delivered to {shipping_address}. (Delivery date not specified).'"
-            elif delivery_date:
-                 user_prompt += f"It was delivered on {delivery_date}. (Shipping address not specified).'"
+            if shipping_address:
+                core_info_parts.append(f"and was delivered to {shipping_address}")
+            else:
+                core_info_parts.append("and was delivered (address not specified)")
+            if delivery_date:
+                core_info_parts.append(f"on {delivery_date}")
             else:
-                 user_prompt += "(Delivery details not fully specified).'"
-        else: # Not delivered or other status
-            user_prompt += f"\nPlease formulate a human-like response to {customer_name}. For example: "
-            user_prompt += f"'Hi {customer_name}, the current status of your order {order_id} for {item_description} is \"{order_status}\".'"
+                core_info_parts.append("(delivery date not specified)")
         
-        user_prompt += f"\n\nFor your reference, the full context originally retrieved from the database was: {context}"
+        core_information_to_include = ", ".join(core_info_parts[:-1]) + (f" {core_info_parts[-1]}" if len(core_info_parts) > 1 else "")
+        if not order_status.lower() == "delivered" and len(core_info_parts) > 1 : # for non-delivered, avoid 'and' before status
+             core_information_to_include = f"your order {order_id} for {item_description} has a status of '{order_status}'"
+
+
+        user_prompt = (
+            f"Customer: {customer_name}\n"
+            f"Order ID: {order_id}\n"
+            f"Item(s): {item_description}\n"
+            f"Status: {order_status}\n"
+        )
+        if order_status.lower() == "delivered":
+            user_prompt += f"Shipping Address: {shipping_address if shipping_address else 'Not specified'}\n"
+            user_prompt += f"Delivered On: {delivery_date if delivery_date else 'Not specified'}\n"
+        
+        user_prompt += f"\nOriginal user query for context: '{query}'\n\n"
+        user_prompt += (
+            f"Your task: Generate a single, complete, and human-like sentence that starts with a greeting to {customer_name}. "
+            f"This sentence MUST convey the following essential information: {core_information_to_include}.\n"
+            f"For example, if all details are present for a delivered order: 'Hi {customer_name}, {core_information_to_include}.'\n"
+            f"For example, for a non-delivered order: 'Hi {customer_name}, {core_information_to_include}.'\n"
+            f"IMPORTANT: Do not ask questions. Do not add any extra conversational fluff. Just provide the single, informative sentence as requested. "
+            f"Ensure the sentence flows naturally and uses the details you've been given.\n"
+            f"Respond now with ONLY that single sentence."
+        )
+        # For LLM's deeper reference, though the primary instruction is above:
+        # user_prompt += f"\n\nFull database context for your reference if needed: {context}"
 
     else: # Default prompt structure for other intents or if details are missing
         system_message = "You are a helpful customer support assistant."
@@ -257,9 +266,12 @@ Assistant Answer:
                 {"role": "system", "content": system_message},
                 {"role": "user", "content": user_prompt}
             ],
-            model=model, temperature=0.7, max_tokens=1024, top_p=1
+            model=model, 
+            temperature=0.5, # Slightly lower temperature might help with stricter adherence
+            max_tokens=1024, 
+            top_p=1
         )
-        response = chat_completion.choices[0].message.content
+        response = chat_completion.choices[0].message.content.strip() # Added strip()
         return response
     except Exception as e:
         st.error(f"Error calling GROQ API: {e}")
@@ -267,7 +279,6 @@ Assistant Answer:
 
 
 def initialize_groq_client(api_key_val):
-    """Initializes the GROQ client."""
     if not api_key_val:
         st.warning("GROQ API Key is missing.")
         return None
@@ -280,7 +291,6 @@ def initialize_groq_client(api_key_val):
 
 # --- Twilio Operations ---
 def initialize_twilio_client(acc_sid, auth_tkn):
-    """Initializes the Twilio client."""
     if not acc_sid or not auth_tkn:
         st.warning("Twilio Account SID or Auth Token is missing.")
         return None
@@ -293,7 +303,6 @@ def initialize_twilio_client(acc_sid, auth_tkn):
 
 def get_new_whatsapp_messages(twilio_client, conversation_service_sid_val, bot_start_time_utc,
                               processed_message_sids, bot_whatsapp_identity_val):
-    """Fetches new, unanswered WhatsApp messages from Twilio Conversations."""
     if not twilio_client:
         st.warning("Twilio client not initialized.")
         return []
@@ -336,7 +345,6 @@ def get_new_whatsapp_messages(twilio_client, conversation_service_sid_val, bot_s
     return sorted(new_messages_to_process, key=lambda m: m['timestamp_utc']) 
 
 def send_whatsapp_message(twilio_client, conversation_service_sid_val, conversation_sid, message_body, bot_identity_val):
-    """Sends a message to a Twilio Conversation from the bot's identity."""
     if not twilio_client:
         st.error("Twilio client not initialized for sending message.")
         return False
@@ -549,13 +557,9 @@ if st.session_state.get("app_started") and st.session_state.get("rag_pipeline_re
 
             context_for_llm, raw_context_data = "No specific context could be retrieved.", None 
             
-            # Parameters for personalized LLM response
-            extracted_customer_name = None
-            extracted_item_name = None
-            extracted_shipping_address = None
-            extracted_delivery_date = None
-            extracted_order_id = None
-            extracted_order_status = None
+            extracted_customer_name, extracted_item_name, extracted_shipping_address, \
+            extracted_delivery_date, extracted_order_id, extracted_order_status = [None] * 6
+
 
             if intent == "ORDER_STATUS":
                 order_id_to_check = None
@@ -568,9 +572,12 @@ if st.session_state.get("app_started") and st.session_state.get("rag_pipeline_re
                 
                 if order_id_to_check:
                     raw_context_data = get_order_details(order_id_to_check, st.session_state.customer_orders_data)
-                    context_for_llm = f"Order Details for {order_id_to_check}: {raw_context_data}"
-                    # Try to parse details for personalized response
-                    if not raw_context_data.startswith("No order found") and not raw_context_data.startswith("Customer order data is not loaded"):
+                    # context_for_llm will be used as the 'context' parameter in generate_response_groq
+                    # For ORDER_STATUS, this raw_context_data (JSON string) is still useful for LLM's reference,
+                    # even though specific fields are extracted for the specialized prompt.
+                    context_for_llm = raw_context_data 
+                    
+                    if isinstance(raw_context_data, str) and not raw_context_data.startswith("No order found") and not raw_context_data.startswith("Customer order data is not loaded"):
                         try:
                             order_data_dict = json.loads(raw_context_data)
                             extracted_customer_name = order_data_dict.get("customer_name")
@@ -578,21 +585,23 @@ if st.session_state.get("app_started") and st.session_state.get("rag_pipeline_re
                             if items and len(items) > 0 and isinstance(items[0], dict):
                                 extracted_item_name = items[0].get("name", "your item(s)")
                             else:
-                                extracted_item_name = "your item(s)"
+                                extracted_item_name = "your item(s)" # Fallback
                             extracted_shipping_address = order_data_dict.get("shipping_address")
                             extracted_delivery_date = order_data_dict.get("delivered_on")
                             extracted_order_status = order_data_dict.get("status")
-                            extracted_order_id = order_data_dict.get("order_id")
+                            extracted_order_id = order_data_dict.get("order_id") # Should be same as order_id_to_check
                         except json.JSONDecodeError:
                             st.warning(f"Could not parse order details JSON for {order_id_to_check} for personalization.")
-                            # raw_context_data already holds the error or JSON string
+                            context_for_llm = f"Error parsing order details for {order_id_to_check}. Raw data: {raw_context_data}"
+                    elif isinstance(raw_context_data, str): # Handle "No order found" or "data not loaded"
+                         context_for_llm = raw_context_data # LLM will state this
                 else:
                     context_for_llm = "To check an order status, please provide a valid Order ID (e.g., ORD123)."
                     raw_context_data = {"message": "Order ID needed or not found in query."}
 
             elif intent == "PRODUCT_INFO":
                 raw_context_data = get_product_info(user_query_manual, st.session_state.products_data)
-                context_for_llm = f"Product Information related to '{user_query_manual}': {raw_context_data}"
+                context_for_llm = raw_context_data # Product info is directly used as context
             
             elif intent == "GENERAL_POLICY_FAQ" or intent == "UNKNOWN": 
                 if st.session_state.faiss_index_pdfs and st.session_state.embedding_model and st.session_state.indexed_pdf_chunks:
@@ -612,13 +621,13 @@ if st.session_state.get("app_started") and st.session_state.get("rag_pipeline_re
             llm_response = generate_response_groq(
                 _groq_client=st.session_state.groq_client, 
                 query=user_query_manual, 
-                context=context_for_llm, # Pass the full context string for LLM's general reference
+                context=context_for_llm, 
                 intent=intent,
                 customer_name=extracted_customer_name,
                 item_name=extracted_item_name,
                 shipping_address=extracted_shipping_address,
                 delivery_date=extracted_delivery_date,
-                order_id=extracted_order_id,
+                order_id=extracted_order_id, # This will be the specific order ID from user query
                 order_status=extracted_order_status
             )
             
@@ -675,9 +684,10 @@ if st.session_state.get("bot_started") and st.session_state.get("rag_pipeline_re
 
                         context_for_llm_whatsapp = "No specific context could be retrieved."
                         raw_context_data_whatsapp = None
+                        
+                        wa_customer_name, wa_item_name, wa_shipping_address, \
+                        wa_delivery_date, wa_order_id, wa_order_status = [None] * 6
 
-                        # Parameters for personalized LLM response
-                        wa_customer_name, wa_item_name, wa_shipping_address, wa_delivery_date, wa_order_id, wa_order_status = [None] * 6
 
                         if intent_whatsapp == "ORDER_STATUS":
                             order_id_to_check_whatsapp = None
@@ -690,8 +700,9 @@ if st.session_state.get("bot_started") and st.session_state.get("rag_pipeline_re
 
                             if order_id_to_check_whatsapp:
                                 raw_context_data_whatsapp = get_order_details(order_id_to_check_whatsapp, st.session_state.customer_orders_data)
-                                context_for_llm_whatsapp = f"Order Details for {order_id_to_check_whatsapp}: {raw_context_data_whatsapp}"
-                                if not raw_context_data_whatsapp.startswith("No order found") and not raw_context_data_whatsapp.startswith("Customer order data is not loaded"):
+                                context_for_llm_whatsapp = raw_context_data_whatsapp # Full JSON string as context
+
+                                if isinstance(raw_context_data_whatsapp, str) and not raw_context_data_whatsapp.startswith("No order found") and not raw_context_data_whatsapp.startswith("Customer order data is not loaded"):
                                     try:
                                         order_data_dict_wa = json.loads(raw_context_data_whatsapp)
                                         wa_customer_name = order_data_dict_wa.get("customer_name")
@@ -706,6 +717,9 @@ if st.session_state.get("bot_started") and st.session_state.get("rag_pipeline_re
                                         wa_order_id = order_data_dict_wa.get("order_id")
                                     except json.JSONDecodeError:
                                          st.warning(f"Could not parse order details JSON for {order_id_to_check_whatsapp} (WhatsApp) for personalization.")
+                                         context_for_llm_whatsapp = f"Error parsing order details for {order_id_to_check_whatsapp}. Raw data: {raw_context_data_whatsapp}"
+                                elif isinstance(raw_context_data_whatsapp, str):
+                                     context_for_llm_whatsapp = raw_context_data_whatsapp
                             else:
                                 context_for_llm_whatsapp = "To check an order status, please provide a valid Order ID (e.g., ORD123)."
                                 raw_context_data_whatsapp = {"message": "Order ID needed or not found in query."}
@@ -713,7 +727,7 @@ if st.session_state.get("bot_started") and st.session_state.get("rag_pipeline_re
 
                         elif intent_whatsapp == "PRODUCT_INFO":
                             raw_context_data_whatsapp = get_product_info(user_query_whatsapp, st.session_state.products_data)
-                            context_for_llm_whatsapp = f"Product Information related to '{user_query_whatsapp}': {raw_context_data_whatsapp}"
+                            context_for_llm_whatsapp = raw_context_data_whatsapp
                         
                         elif intent_whatsapp == "GENERAL_POLICY_FAQ" or intent_whatsapp == "UNKNOWN":
                             if st.session_state.faiss_index_pdfs and st.session_state.embedding_model and st.session_state.indexed_pdf_chunks: