At Google Cloud Workspace, I shipped a production LLM agent that helps enterprise administrators investigate security incidents through natural-language queries, typed parameters, and schema-valid RPC calls.
I fine-tuned Gemini across 15 investigation scenarios and co-authored the evaluation dataset used to gate model rollouts. I also owned an autorater for end-to-end tool selection, improving accuracy from 85% to 99%, and used SQL A/B analysis to identify variants that lifted upgrade conversions by 40%.
Previously, I built distributed firmware-simulation infrastructure and high-throughput Java services at Google, including an email-verification pipeline scaled from 5 to 10,000 QPS. I also build distributed systems independently, from a FastAPI collaboration backend to a YOLO-based road-inspection pipeline.
