bias_guardrail.py

The article explains how to add real-time safety and bias guardrails to a generative AI career advisor.

It combines prompt sanitization, token-level checks, post-generation moderation, and audit logging in a serverless setup.

A Python bias-detection component scores outputs against protected attributes to help monitor fairness and compliance.

The approach is designed to reduce harmful or privacy-risky advice while supporting the UK Online Safety Act and the EU Digital Services Act.