I turn ambiguous quality into clear, defensible metrics.
Tech lead for agentic evaluation on Gemini in Chrome — building the infrastructure that measures what AI systems can actually do, reliably, at scale. Eight years at Google, from Chromium internals relied on by billions to the eval platforms that decide what ships.
I lead model evaluation and agentic quality for Gemini in Chrome, where my job is to turn fuzzy questions about capability and reliability into signals a team can trust: deterministic offline replay so results are reproducible, autoraters calibrated against human judgment, distributed eval pipelines with health dashboards and on-call alerting, and the observability tooling used to root-cause anomalous results.
Before that, three years deep in Chromium internals on architecture relied on by an estimated two billion people, and research in multi-agent systems and multi-objective optimization — published work and a granted patent.
I grew up in Hyderabad in an entrepreneurial family that shaped how I lead: growth mindset, resilience, service to community. Tennis and cricket since age ten taught me teamwork; Vipassana clarified the purpose underneath it — build technology that makes the world more equitable. As an angel investor, I back founders who share that conviction.
Defensible metrics, not vibes.
Mountain View, CA
- Define what "good" means for Chrome's Gemini-powered web agents across reasoning and multi-step behavior, set the correctness bars the team ships against, and drive the eval roadmap for 5+ senior engineers.
- Raised first-party tool correctness 50% → 94% on stateful eval infrastructure — dynamic state seeding, isolated concurrent runs, lifecycle management — now adopted by sister agent teams.
- Built deterministic trajectory capture and offline replay, eliminating the flakiness that had been blocking launch decisions and separating real model gains from noise.
- Improved in-house autorater confidence 60% → 90% by analyzing human-vs-model rating disagreements — saved millions in rater spend.
- Built and operate the continuous evaluation platform and the observability layer used to root-cause anomalous results — −70% mean-time-to-debug for multi-step failures.
Seattle, WA
- Built the user-feedback logging pipeline that remains Chrome modeling teams' primary loop for LLM iteration — real usage into training and eval signal at Chrome scale.
- Shipped 100+ ms of page-load latency wins on Chrome's critical path, with measurable revenue impact across Search and Ads.
- Designed the SpareTab navigation architecture and built Precog and Slackline — infrastructure for data-driven performance ship decisions on Perfetto.
London, UK
- Shipped Multi-Page Architecture (MPArch) — the foundational refactor that unblocked Prerendering, BackForward Cache, Portals, and Fenced Frames across the browser.
- Designed and launched BackForward Cache, accelerating back/forward navigation by 1,000 ms for an estimated two billion Chrome users.
- Founded and have run Chromium's GSoC program for six years, coordinating 200+ mentors, admins, and contributors worldwide.
- Shipped 65 of 65 student projects upstream — a 100% completion rate across every cohort.
Rigor over vibes
If it matters, it should be measurable and defensible. I'd rather build the metric than argue about the anecdote.
Honesty & integrity
Absolute truth in the architecture and in the team. Good systems and good cultures both depend on it.
Compassion & growth
Empathy and constant learning, bringing out the best in the people around me. Service to community is the point.
- 2026 Google Tech Impact Award — recognized by senior leadership for technical innovation and cross-team collaboration on Gemini in Chrome.
- 2023 Speaker, Grace Hopper Celebration.
- — Individual recognition from Chrome VP Parisa Tabriz and SVP Hiroshi Lockheimer.
- — Google Venkat Scholar · Get-Ahead Scholar.