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
+from datetime import datetime
+import json
 
 # === Configuration ===
 genai.configure(api_key=os.environ["GEMINI_API_KEY"])
@@ -14,6 +18,40 @@ embedding_model = "models/embedding-001"
 llm_model_name = "models/gemma-3-4b-it"
 collection_name = "xeno_collection"
 
+# === Google Sheets Setup for Hugging Face ===
+# Use environment variable for Google Sheets credentials
+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 (replace 'Response_Log' with your Google Sheet name)
+sheet = client_gspread.open("Response_Log").sheet1
+
+def log_response(question, answer, source_ids):
+    """
+    Log a question, answer, and source IDs 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.
+    """
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    row = [timestamp, question, answer, source_ids]
+    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}")
+
 # === Load and Clean Knowledge Base ===
 df_kb = pd.read_json("XENO_Uganda_KnowledgeBase_Advisory.json")
 df_kb.dropna(subset=['Content'], inplace=True)
@@ -28,7 +66,8 @@ 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"]  # Ensure ID is included in metadata
         })
         ids.append(item["ID"])
     return documents, metadatas, ids
@@ -37,7 +76,7 @@ xeno_data_list = df_kb.to_dict('records')
 documents, metadatas, ids = prepare_documents(xeno_data_list)
 
 # === Setup ChromaDB ===
-client = chromadb.PersistentClient(path="./xeno_db")
+client = chromadb.PersistentClient(path="/tmp/xeno_db")  # Use /tmp for Hugging Face Spaces
 try:
     collection = client.get_collection(name=collection_name)
 except:
@@ -112,6 +151,7 @@ Remember: This is a single-turn interaction. You have no memory of previous conv
 def process_context(results, cosine_scores, max_results=2):
     sorted_indices = np.argsort(cosine_scores)[::-1][:max_results]
     formatted_context = ""
+    source_ids = []
     for i, idx in enumerate(sorted_indices, 1):
         result = results[idx]
         score = cosine_scores[idx]
@@ -119,7 +159,8 @@ def process_context(results, cosine_scores, max_results=2):
         formatted_context += f"Q: {result.metadata.get('question', 'N/A')}\n"
         formatted_context += f"A: {result.metadata.get('content', 'N/A')}\n"
         formatted_context += "-" * 40 + "\n"
-    return formatted_context
+        source_ids.append(result.metadata.get('id', 'N/A'))
+    return formatted_context, source_ids
 
 # === LLM Generation ===
 def generate_xeno_response(context, question):
@@ -137,6 +178,7 @@ def generate_xeno_response(context, question):
 # === Main Interface Logic ===
 def get_context_and_answer(message, history):
     if message.lower().strip() in {"hi", "hello", "hey"}:
+        log_response(message, "Hello! How can I assist you with XENO services today?", "N/A")
         return "Hello! How can I assist you with XENO services today?"
 
     queried_results = retriever.invoke(message)
@@ -153,10 +195,13 @@ def get_context_and_answer(message, history):
 
     # If none of the results have sufficient similarity, fallback
     if max(cosine_scores) < 0.4:
+        log_response(message, "I'm sorry, I couldn't find the specific information you're looking for in my knowledge base.", "N/A")
         return "I'm sorry, I couldn't find the specific information you're looking for in my knowledge base."
 
-    context = process_context(queried_results, cosine_scores)
-    return generate_xeno_response(context, message)
+    context, source_ids = process_context(queried_results, cosine_scores)
+    answer = generate_xeno_response(context, message)
+    log_response(message, answer, ", ".join(source_ids))
+    return answer
 
 # === Gradio UI ===
 iface = gr.ChatInterface(
@@ -167,4 +212,4 @@ iface = gr.ChatInterface(
 )
 
 if __name__ == "__main__":
-    iface.launch()
+    iface.launch(share=False)  # Set share=False for Hugging Face Spaces