At SportsBiz, I lead the AI Center of Excellence, setting architecture standards, evaluation gates, and delivery expectations for 6–10 engineers plus external contractors. I also own a production vision-language pipeline on Amazon Bedrock, where few-shot patterns raised Post Pad / Field Stake Flag accuracy from 58–64% to ~86% on 517 human-labeled frames.
At Alstom, I owned a production LangGraph multi-agent debugging platform for rail-maintenance HPC systems. I defined hybrid retrieval, citation-required generation, and a human gate before action-changing steps; gold-suite quality moved from ~65% to ~90%.
At DeepIntel, I built a TypeScript/Next.js and Claude agent platform on Azure AI Foundry, with a Strategy Oversight Agent and human-in-the-loop Acknowledge / Escalate / Regenerate controls. Earlier, at Westinghouse Electric Company and FS-Elliott, I worked on HPC testing, simulation tools, and AI automation frameworks.
