Amara AI
service : product design
timeline : January 2024
role : data scientist, ML engineer
Amara AI is a credit risk assessment platform designed to provide transparent and explainable credit scoring for financial institutions.
Traditional credit scoring models were opaque and excluded thin-file borrowers who lacked extensive credit history. Our goal was to create a more inclusive system while maintaining high predictive accuracy.
problem
Financial institutions needed a way to assess creditworthiness for borrowers with limited credit history while maintaining transparency for regulatory compliance.
approach
Developed an ensemble model combining XGBoost and LightGBM, incorporating alternative data sources including transaction patterns and behavioral signals. Implemented SHAP (SHapley Additive exPlanations) values to provide feature-level explanations for each prediction.
outcome
Improved Gini coefficient by 15% compared to the previous model. Reduced false positives by 8%, leading to better approval rates for qualified borrowers while maintaining risk thresholds.