I built a credit risk scoring platform that serves risk tiers, default probabilities, credit scores, and feature importances through a FastAPI REST API. I containerized it with Docker, deployed it on AWS EC2, and stored model artifacts in S3.
For the platform, I added a GitHub Actions CI/CD pipeline and replaced pickle inference with a NumPy-only implementation, verifying it against the original. I also corrected probability inflation through reliability-curve calibration and selected Gradient Boosting for production based on interpretability.
My other projects include a financial NLP pipeline that reached Macro F1 0.96, and a multi-factor research and derivatives-pricing project. In my current part-time role as an Accounts & Audit Assistant, I analyze financial statements and perform GST filings, TDS computations, and audit procedures.
