Update app.py

app.py

@@ -7,6 +7,10 @@ from sentence_transformers import util
 import google.generativeai as genai
 import chromadb
 from langchain_chroma import Chroma
+import gspread
+from google.oauth2.service_account import Credentials
+import json
+from datetime import datetime
 import re
 from typing import Dict, List, Tuple
 
@@ -16,6 +20,55 @@ embedding_model = "models/embedding-001"
 llm_model_name = "models/gemma-3-4b-it"
 collection_name = "xeno_collection"
 
+# === Google Sheets Setup for Hugging Face ===
+def get_google_sheets_credentials():
+    credentials_json = os.environ.get("GOOGLE_SHEETS_CREDENTIALS")
+    if not credentials_json:
+        raise ValueError("GOOGLE_SHEETS_CREDENTIALS environment variable not set.")
+    credentials_dict = json.loads(credentials_json)
+    scope = ["https://spreadsheets.google.com/feeds", "https://www.googleapis.com/auth/drive"]
+    creds = Credentials.from_service_account_info(credentials_dict, scopes=scope)
+    return creds
+
+# Authenticate with Google Sheets
+client_gspread = gspread.authorize(get_google_sheets_credentials())
+
+# Open the Google Sheet
+sheet = client_gspread.open("Response_Log").sheet1
+
+def log_response(question, answer, source_ids, knowledge_pairs):
+    """
+    Log a question, answer, source IDs, and knowledge base question-answer pairs to the Google Sheet.
+    
+    Args:
+        question (str): The question asked by the user.
+        answer (str): The answer provided by the model.
+        source_ids (str): Comma-separated list of source IDs used.
+        knowledge_pairs (list): List of tuples containing (question, answer) from the knowledge base.
+    """
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    knowledge_question_1 = knowledge_pairs[0][0] if len(knowledge_pairs) > 0 else "N/A"
+    knowledge_answer_1 = knowledge_pairs[0][1] if len(knowledge_pairs) > 0 else "N/A"
+    knowledge_question_2 = knowledge_pairs[1][0] if len(knowledge_pairs) > 1 else "N/A"
+    knowledge_answer_2 = knowledge_pairs[1][1] if len(knowledge_pairs) > 1 else "N/A"
+    row = [
+        timestamp,
+        question,
+        answer,
+        source_ids,
+        knowledge_question_1,
+        knowledge_answer_1,
+        knowledge_question_2,
+        knowledge_answer_2
+    ]
+    try:
+        sheet.append_row(row)
+        print(f"Logged: {question} | Source IDs: {source_ids}")
+    except Exception as e:
+        print(f"Failed to log to Google Sheet: {e}")
+        with open("/tmp/response_log.txt", "a") as f:
+            f.write(f"{timestamp},{question},{answer},{source_ids},{knowledge_question_1},{knowledge_answer_1},{knowledge_question_2},{knowledge_answer_2}\n")
+
 # === Intent Classification System ===
 class IntentClassifier:
     def __init__(self):
@@ -66,16 +119,13 @@ class IntentClassifier:
         """
         message_lower = message.lower().strip()
         
-        # Check for each intent pattern
         for intent_name, intent_data in self.intent_patterns.items():
             for pattern in intent_data['patterns']:
                 if re.search(pattern, message_lower, re.IGNORECASE):
-                    # Return random response from available responses
                     import random
                     response = random.choice(intent_data['responses'])
                     return intent_name, response
         
-        # If no simple intent found, it's a query that needs RAG
         return 'query', ''
     
     def is_simple_intent(self, intent: str) -> bool:
@@ -100,7 +150,8 @@ def prepare_documents(data):
             "section": item.get("Section", ""),
             "source": item.get("Source", ""),
             "owner": item.get("Owner", ""),
-            "tag": item.get("Tag", "")
+            "tag": item.get("Tag", ""),
+            "id": item["ID"]
         })
         ids.append(item["ID"])
     return documents, metadatas, ids
@@ -109,12 +160,18 @@ xeno_data_list = df_kb.to_dict('records')
 documents, metadatas, ids = prepare_documents(xeno_data_list)
 
 # === Setup ChromaDB ===
-client = chromadb.PersistentClient(path="./xeno_db")
 try:
-    collection = client.get_collection(name=collection_name)
-except:
-    collection = client.create_collection(name=collection_name)
-    collection.add(documents=documents, metadatas=metadatas, ids=ids)
+    client = chromadb.PersistentClient(path="/tmp/xeno_db")
+    try:
+        collection = client.get_collection(name=collection_name)
+        print(f"Loaded existing ChromaDB collection: {collection_name}")
+    except:
+        print(f"Creating new ChromaDB collection: {collection_name}")
+        collection = client.create_collection(name=collection_name)
+        collection.add(documents=documents, metadatas=metadatas, ids=ids)
+except Exception as e:
+    print(f"Failed to initialize ChromaDB: {e}")
+    raise
 
 vector_store = Chroma(client=client, collection_name=collection_name)
 retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 4})
@@ -131,14 +188,20 @@ remember previous conversations."""
 def process_context(results, cosine_scores, max_results=2):
     sorted_indices = np.argsort(cosine_scores)[::-1][:max_results]
     formatted_context = ""
+    source_ids = []
+    knowledge_pairs = []
     for i, idx in enumerate(sorted_indices, 1):
         result = results[idx]
         score = cosine_scores[idx]
+        question = result.metadata.get('question', 'N/A')
+        answer = result.metadata.get('content', 'N/A')
         formatted_context += f"Knowledge Entry {i}:\n"
-        formatted_context += f"Q: {result.metadata.get('question', 'N/A')}\n"
-        formatted_context += f"A: {result.metadata.get('content', 'N/A')}\n"
+        formatted_context += f"Q: {question}\n"
+        formatted_context += f"A: {answer}\n"
         formatted_context += "-" * 40 + "\n"
-    return formatted_context
+        source_ids.append(result.metadata.get('id', 'N/A'))
+        knowledge_pairs.append((question, answer))
+    return formatted_context, source_ids, knowledge_pairs
 
 # === LLM Generation ===
 def generate_xeno_response(context, question):
@@ -161,13 +224,16 @@ def get_context_and_answer(message, history):
     
     # Step 2: Handle simple intents directly
     if intent_classifier.is_simple_intent(intent) and direct_response:
+        log_response(message, direct_response, "N/A", [])
         return direct_response
     
     # Step 3: For queries that need RAG processing
     if intent == 'query':
         # Check if message is too short or unclear
         if len(message.strip()) < 3:
-            return "I'd be happy to help! Could you please provide more details about what you'd like to know about XENO services?"
+            answer = "I'd be happy to help! Could you please provide more details about what you'd like to know about XENO services?"
+            log_response(message, answer, "N/A", [])
+            return answer
         
         # Retrieve relevant documents
         try:
@@ -193,16 +259,29 @@ def get_context_and_answer(message, history):
 
             # If none of the results have sufficient similarity, fallback
             if max(cosine_scores) < 0.4:
-                return "I'm sorry, I couldn't find the specific information you're looking for in my knowledge base. Could you try rephrasing your question or contact XENO support directly for assistance?"
+                answer = "I'm sorry, I couldn't find the specific information you're looking for in my knowledge base. Could you try rephrasing your question or contact XENO support directly for assistance?"
+                log_response(message, answer, "N/A", [])
+                return answer
 
-            context = process_context(queried_results, cosine_scores)
-            return generate_xeno_response(context, message)
+            context, source_ids, knowledge_pairs = process_context(queried_results, cosine_scores)
+            answer = generate_xeno_response(context, message)
+            log_response(message, answer, ", ".join(source_ids), knowledge_pairs)
+            return answer
             
         except Exception as e:
-            return "I apologize, but I'm experiencing a technical issue. Please contact XENO support directly for assistance with your query."
+            answer = "I apologize, but I'm experiencing a technical issue. Please contact XENO support directly for assistance with your query."
+            log_response(message, answer, "N/A", [])
+            return answer
     
+    # Handle goodbye intent (not simple, but has direct response)
+    if intent == 'goodbye' and direct_response:
+        log_response(message, direct_response, "N/A", [])
+        return direct_response
+
     # Fallback for any unhandled cases
-    return "I'm here to help with XENO financial services. What would you like to know?"
+    answer = "I'm here to help with XENO financial services. What would you like to know?"
+    log_response(message, answer, "N/A", [])
+    return answer
 
 # === Enhanced Gradio UI ===
 def create_interface():
@@ -226,4 +305,4 @@ I can help you with questions about XENO financial services including:
 # === Main Execution ===
 if __name__ == "__main__":
     iface = create_interface()
-    iface.launch()
+    iface.launch(share=False)