LLM-powered tool to detect PII in logs for privacy and GDPR compliance (for fun)

Overview

PII Guard is an LLM-powered tool that detects and manages Personally Identifiable Information (PII) in logs, and is described as designed to support data privacy and GDPR compliance. The repository states that it is a personal side project, built to explore how Large Language Models can detect sensitive data in logs more intelligently than traditional regex-based approaches. The project experiments with Large Language Models, specifically the gemma:3b model running locally via Ollama, to evaluate how effectively they can identify PII in both structured and unstructured log data. According to the README, it analyzes logs using natural language understanding, handles real-world, messy logs better than regex, and identifies PII even when it is obfuscated, incomplete, or embedded in text. It is also said to handle multilingual input and inconsistent formats and to leverage semantic context instead of relying on static patterns. The README lists the PII types detected across several groups: identity information (such as full-name, email, phone-number and address), sensitive categories under GDPR Art. 9, government and financial identifiers, network and device information, and vehicle information such as license-plate. The tool is started with a single command, make all-in-up, which launches a full stack including PostgreSQL, Elasticsearch, RabbitMQ, Ollama with gemma:3b, and the PII Guard dashboard and backend API. A web interface is available at http://localhost:3000 and an API endpoint at http://localhost:8888/api/jobs. The README describes the project as a work in progress and welcomes contributions.