At Samsung, I build and own Connected TV impression-capacity forecasting that supports delivery guarantees for campaigns representing 15% of ad revenue. My projections help 98% of campaigns meet impression goals.
I productionized two-stage inventory forecasting with Temporal Fusion Transformer and AutoGluon, automated retraining and drift monitoring, and reduced model refresh time by 83%. I also built CrewAI and LangChain automation that cut media-planning time by 85% and report preparation by 95%.
Previously, I built data observability and configuration services for Samsung’s unified data platform, managing roughly 1 TB per day of ingestion. I’ve also mentored a PRISM team on multimodal ad-creative detection and published research in explainable parse trees and event extraction.
