I built the Regulatory Change-Impact Agent, a FastAPI service that uses LLMs and hybrid BM25 + TF-IDF retrieval to connect regulatory text with affected policies. On a hand-labelled evaluation set, retrieval reached 88.2% Recall@5, and the service generates evidence-cited redline suggestions for human review.
For the Automated Control Testing Simulator, I built a Python pipeline that ran eight financial control tests on synthetic transactions. The rule tests reached 91.1% macro F1 on injected violations, and an Isolation Forest flagged additional transactions for review.
I also built CardioFederate, a Flower-based federated learning pipeline for ECG reports, and provisioned its GCP infrastructure with Terraform. At Vinyas Innovative Technologies Limited, I contributed to understanding sensor-data monitoring needs for an IoT-based blood monitoring system.

