Urvesh Patel

Urvesh works on voice AI quality and platform engineering at Eve. He built the team's eval harness and Voice Agent Bench, which freezes real production calls at the moment they broke and replays them against every prompt or model change before it ships — the subject of his post The Eval Harness Behind Our Voice Agent. He designs and runs the team's live experiments on the voice agent, from model swaps to pacing tuning, and the from-number caller-trust system that lets the agent recognize returning callers from call history alone.

Before voice, he worked on document intelligence for litigation: the medical-chronology pipeline that extracts and links billing records from medical records, an exhibit-selection system that matches case documents (including scanned images, via generated visual summaries) to draft citations, and an early citation-verification system since replaced by newer infrastructure.