BloombergGraphics/2024-openai-gpt-hiring-racial-discrimination

Implements a controlled resume-ranking experiment using OpenAI's Chat API and embedding models (GPT-3.5/4 and ADA-002) to quantify hiring discrimination across demographic groups. The workflow chains four sequential Jupyter notebooks that derive demographically-distinct names from Census and voter registration data, systematically rank near-identical resumes by job title, perform statistical hypothesis testing for bias signals, and visualize name embeddings in 2D space using UMAP dimensionality reduction. All pipeline artifacts—from input name datasets to intermediate API responses to aggregated performance metrics—are versioned and reproducible via OpenAI API credentials.

No commits in the last 6 months.