AI Powered Auction Arbitrage Agent

Designed and deployed a system that prices auction lots and converts each into a max-bid recommendation against an ROI target — 2,800 lots across two live deployments. A Python pipeline ingests listings and resolves market comps through a three-source confidence-weighted lookup, with entity matching on unstructured lot titles handled by GPU-accelerated OCR and an embedding engine. Encodes buyer’s fees and bid multipliers for 18 auction houses, and persists results in AWS (DynamoDB, S3) for backtesting and future model training.