dreji18/Fairness-in-AI

Provides end-to-end NLP bias detection and mitigation through four specialized capabilities: classification, word/phrase recognition, masking, and text debiasing. Built on an ensemble of three Transformer models (DistilBERT for classification, RoBERTa for entity recognition, and a masked language model for text generation), trained on the MBIC dataset for news article analysis. Installable via pip and designed for production use on raw text requiring preprocessing and bias probability scoring.

102 stars. No commits in the last 6 months.