At DV Data & Analytics, I build production data science systems with leakage checks, recorded promotion decisions, and verified train/serve parity. My work includes payment fraud scoring, demand forecasting, delivery-time prediction, and customer segmentation.
For payment fraud scoring, I identified and corrected features that exposed post-payment information, then evaluated the model with design-weighted metrics to account for partially observed labels. The system runs on AWS EKS, with an immutable model registry and automated rollback.
I also built a pooled chocolate-demand forecast across 18 brands and four zones, and developed a delivery-time model and quick-commerce customer segmentation system. Across these projects, I’ve focused on rigorous evaluation, production deployment, and making model failures visible rather than silent.
