PII Guard is an innovative tool utilizing Large Language Models (LLMs) to detect and manage Personally Identifiable Information (PII) within logs, effectively supporting both data privacy initiatives and compliance with GDPR.

Project Overview

Developed as an exploratory personal project, PII Guard leverages the capabilities of the gemma:3b model, operated through Ollama, to intelligently identify PII in various log formats—both structured and unstructured. This approach enhances traditional detection methods, offering improved accuracy over conventional regex-based systems.

Key Features

• Advanced PII Detection: Successfully identifies PII, even when data is obfuscated, incomplete, or embedded in extended text.
• Multilingual Support: Capable of handling diverse languages and inconsistent data formats.
• Semantic Understanding: Utilizes the contextual understanding of language, moving beyond the limitations of static detection patterns.
• Foundation for Privacy Tooling: Serves as a robust platform for developing AI-powered privacy applications.

Types of PII Detected

PII Guard encompasses a wide range of sensitive information:

• Identity Information: full-name, first-name, last-name, username, email, phone-number, address, etc.
• Sensitive Categories (GDPR Art. 9): Covers racial-or-ethnic-origin, political-opinion, religious-belief, health-data, and more.
• Government & Financial Identifiers: Includes national-id, passport-number, credit-card, bank-account, etc.
• Network & Device Information: Detects identifiers such as ip-address, mac-address, browser-fingerprint.
• Vehicle Information: Recognizes license-plate data.

Getting Started

PII Guard is easy to set up. Launch the complete stack using a simple command:

make all-in-up

Shutting down the stack is similarly straightforward:

make all-in-down

This setup includes essential services such as PostgreSQL, Elasticsearch, RabbitMQ, and the PII Guard dashboard.

Testing and API Usage

A web interface is available to interact with the tool at http://localhost:3000, along with an API endpoint located at http://localhost:8888/api/jobs. Sample logs can be submitted via cURL to test the PII detection functionalities.

Project Structure

The repository is organized into several key components, including APIs, dashboards, and LLM prompt templates, available through clear directory links within the project.

Community Engagement

Contributions, feedback, and suggestions for improvements are highly encouraged. Bug reports and feature requests are vital for the ongoing enhancement of the PII Guard tool.