At DoorDash, I lead a four-engineer workstream building the data pipelines behind RAG-based consumer and Dasher assistants. I reduced embedding compute by 63% across about 18 million catalog items, and improved hybrid-retrieval Recall@10 from 0.71 to 0.86.
I also developed data pipelines for an LLM evaluation framework that lowered unsupported answers by 28%, and built telemetry pipelines that helped reduce cost per conversation by 22%. I authored chunking and embedding standards adopted across GenAI teams.
At Cohere Health, I built healthcare data systems for AI-driven prior authorization, including FHIR R4 and HL7 v2 ingestion and governed clinical data. Earlier at Uber, I engineered streaming pipelines processing about 2.5 billion events daily; at Smartsheet, I moved ETL and reporting to Amazon Redshift, reducing report latency from 26 hours to 3 hours.
