Fairness API

The Fairness API is a modular software package written in Python and designed to support fair, transparent, and explainable human resources (HR) decision-making. The system enables organizations to implement and monitor fair ranking algorithms, provide actionable explainability for AI-driven decisions, and assess fairness risks across diverse HR data and processes.

Fairness Monitoring

• Supports multi-party privacy levels: strict two-party, pseudo two-party, and one-party computations.
• Implements fairness metrics: group exposure, top-k fairness, demographic parity, equal opportunity, and intersectional fairness.
• Front-end protected attribute collection to safeguard sensitive data.
• Metric reliability testing for datasets with limited protected attribute data.

Fair Ranking Interventions

• Optimizes rankings for multiple and non-binary protected attributes.
• Implements gFair, a novel group fairness intervention extending iFair.
• Supports intersectional groups while maintaining candidate similarity within embeddings.

Explainability Module

• Counterfactual explanations based on a modified DiCE framework compatible with scikit-learn and LightGBM.
• Factual explainability using ranking_shap for interpreting ranking outcomes.
• Expanded documentation and interpretable-by-design examples using ExplainableBoostingRegressor.

Documentation & Examples

• Includes Jupyter notebooks demonstrating preprocessing, fair ranking interventions, fairness monitoring, and explainability workflows.
• Comprehensive API documentation with clear installation instructions and example pipelines.
• Full documentation available in the deliverable package (GitHub).