PII Anonymization Gateway

Protect sensitive personal information before sending prompts to LLMs like ChatGPT, Claude, or Gemini.

What Does This Do?

This application demonstrates how to safely use Large Language Models (LLMs) with sensitive data by:

1. Detecting PII - Automatically finds names, emails, phone numbers, SSN, Indian IDs (PAN, Aadhaar, etc.)
2. Anonymizing - Replaces PII with placeholders like [PERSON_1], [EMAIL_ADDRESS_1]
3. Safe LLM Queries - Sends anonymized text to Gemini (the LLM never sees real data!)
4. De-anonymizing - Restores original PII in the LLM's response

Real-World Use Cases

• Customer Support: Summarize complaints without exposing customer details
• HR/Recruiting: Draft responses while protecting applicant information
• Healthcare: Analyze patient notes without violating HIPAA
• Legal: Review contracts without exposing party identities
• Finance: Investigate transactions without leaking account details
• Education: Process student data securely

Try It Out

1. Go to the " Chat with LLM" tab
2. Enter a message with personal information
3. Watch how your data is protected in real-time!
4. See the full anonymization → LLM → de-anonymization flow

Technology Stack

• PII Detection: Microsoft Presidio
• NLP Model: spaCy (en_core_web_lg)
• LLM: Google Gemini (via OpenAI-compatible API)
• Frontend: Gradio
• Backend: Python + FastAPI

Supported PII Types

Standard: Names, Emails, Phone Numbers, Credit Cards, SSN, Dates, Locations, IP Addresses

Indian: PAN, Aadhaar, IFSC, GST, Voter ID, Passport, Driving License, Bank Accounts

Note

This is a portfolio demo project. The LLM chat feature uses Gemini's free tier API. For production use, implement proper rate limiting, Redis for session storage, and enterprise API keys.

Links

• Hugging-Face: PII-Anonymization
• LinkedIn: Vikas U Rao

---

Built by Vikas U Rao