Update app.py

app.py

@@ -9,6 +9,7 @@ from llama_index.embeddings.openai import OpenAIEmbedding
 from llama_index.core.postprocessor import MetadataReplacementPostProcessor
 from llama_index.core.node_parser import SentenceWindowNodeParser
 from dotenv import load_dotenv
+import re
 
 # Load environment variables
 load_dotenv("config.env")
@@ -77,6 +78,18 @@ Settings.node_parser = sentence_node_parser
 # Create index
 index = VectorStoreIndex.from_documents(documents)
 
+# Function to filter PII from text (removes emails, phone numbers, etc.)
+def filter_pii(text):
+    # Regular expressions for email, phone numbers, and sensitive patterns
+    email_pattern = r"\S+@\S+\.\S+"
+    phone_pattern = r"\+?\d[\d\s()-]{8,}\d"
+    
+    # Replace found patterns with [REDACTED]
+    text = re.sub(email_pattern, "[REDACTED]", text)
+    text = re.sub(phone_pattern, "[REDACTED]", text)
+    
+    return text
+
 # Custom reranking function
 def custom_rerank(nodes, query):
     rerank_prompt = (
@@ -108,7 +121,7 @@ query_engine = index.as_query_engine(
     ],
 )
 
-# Chatbot function
+# Chatbot function with PII filter
 def chatbot(message, history):
     history_text = "\n".join([f"Human: {h[0]}\nAI: {h[1]}" for h in history])
     full_query = f"Given the following chat history:\n{history_text}\n\nHuman: {message}\nAI:"
@@ -121,20 +134,29 @@ def chatbot(message, history):
     
     # Synthesize answer from reranked nodes
     context = "\n".join([node.get_content() for node in reranked_nodes])
+    
+    # Get the response from LLM
     response = llm.complete(
         f"Using the following context, answer the query:\n\nContext: {context}\n\nQuery: {full_query}"
     )
     
-    return response.text
+    # Filter PII from the response
+    filtered_response = filter_pii(response.text)
+    
+    return filtered_response
 
-# Create Gradio interface
+# Create Gradio interface with comprehensive instructions
 iface = gr.ChatInterface(
     chatbot,
-    title="Resume Chatbot",
-    description="Ask questions about resumes in the database.",
+    title="Resume Chatbot - Secure Candidate Query",
+    description=(
+        "This is a Resume Chatbot that answers questions about candidate experience and qualifications. "
+        "It will not reveal any private information beyond names of candidates. Please ask questions about skills, "
+        "experience, or qualifications without requesting sensitive personal information."
+    ),
     theme="soft",
     examples=[
-        "Out of all the resumes tell me three of them who have experience in SQL?",
+        "Out of all the resumes, tell me three of them who have experience in SQL?",
         "Give me key summary takeaways of the resumes who have experience in Project Management?",
         "Give me the names of 10 candidates who have more than two years of experience in general?",
     ],