At Prospect 33 LLC, I took a trade and transaction reporting detection pipeline from prototype to production for a Tier-1 Global Investment Bank. It applied unsupervised learning across 100+ fields and 16 data streams, and helped confirm more than 57,000 genuine defects in records that had passed existing controls.
I paired SHAP attribution with counterfactual explanations so non-technical reviewers could see what drove a result and what change would reverse it. I also prepared model-ready datasets from fixed-width reporting files and built a configuration-driven pipeline to serve multiple data streams.
For a Top-Tier Global Asset Manager, I ran quarterly monitoring of live demand-forecasting and cost models and onboarded a new forecasting model into the monitoring framework. I also resolved a cascaded join issue during a Hadoop migration and reconciled results against the legacy platform.
Earlier, I built and deployed a Random Forest model for gestational diabetes risk with a Dockerized pipeline and a Streamlit interface. I have a PhD in Electrical Engineering, six years of university teaching experience, and 12 peer-reviewed publications.
