[UofT] PIIGuard: A Hybrid AI pipeline for detecting and redacting your text

Our project, PIIGuard, is a local privacy oriented tool designed to help users safely share text by removing sensitive personal information before it leaves their device. People often copy and paste content into emails, chat platforms, collaborative documents, or AI tools without realizing how much personally identifiable information it may contain. Additionally, many existing PII removal tools rely on cloud processing, which can expose the data they’re meant to protect. PIIGuard focuses on building a lightweight system that performs PII detection entirely on the user’s device using a hybrid approach that combines regular expressions, named entity recognition models, and optionally a small local language model. These methods identify sensitive information such as names, phone numbers, email addresses, identification numbers, and addresses. Detected items are highlighted with labels so users can clearly see what information may be sensitive. The system then generates a sanitized version of the text where detected items are replaced with consistent placeholders, allowing the text to be shared while preserving its structure and meaning. Users paste text into the interface, review detected PII, and copy a sanitized version through a simple and transparent workflow.