BiasLens: An End-to-End Pipeline for Detecting and Mitigating Bias in Vietnamese AI-Generated Text

As AI chatbots, writing assistants, and moderation tools become part of everyday life, more of what people read online is written by machines rather than humans. In Vietnamese, a language with far less AI research and tooling behind it than English, these systems can quietly produce content that stereotypes or demeans particular groups — and because almost no tools exist to check Vietnamese text for this kind of bias, it often goes unnoticed.

BiasLens is our answer to that gap. It's an end-to-end system that reads AI-generated Vietnamese text, flags it across a wide range of bias categories such as gender, age, occupation, religion, and ethnicity, and then automatically rewrites the flagged parts into fairer, more neutral language. Everything is wrapped in a simple web interface, so a developer can paste in a piece of text and, in a few clicks, see where it falls short and get a cleaned-up version back.

Our goal is to give developers building Vietnamese-language AI products a practical way to catch and fix bias before it ever reaches real users.