At Amazon Prime Video, I built a greenfield pipeline to discover shoppable frames across 91,000 titles. I cut LLM calls per video from 2,400 to 60 and devised a CLIP-based metric to score how shoppable each frame is.
At ShopUp, I built a Python and FastMCP server that lets employees fetch reports and metrics through Claude. It was rolled out company-wide, helped retain a client, and included safeguards for database queries.
My work also includes applying YOLO and Hugging Face models to aerial-video scene understanding at The ePlane Co., IIT Madras, and building a supply-chain chatbot with SageMaker and LSTM. I’ve also co-authored papers on multi-agent LLM frameworks and generative AI for supply chains.
