Gemini Infrastructure in Google DeepMind: core contributor to a bulk inference tool for Gemini Large Language Models (LLMs). Before that, I designed and implemented t5bulk, a bulk inference tool for t5x, a widely-used family of LLMs. t5bulk enabled 100s of launches across the company; it was so influential to t5x's success, I'm listed as an author on the t5x paper :)

Pre-LLM NLP in Google Research: Compact neural networks for on-device NLP. Used in Android and in the Chrome browser (1B+ installs). Main engineer, working with a team of Google researchers led by David Weiss

Speech Team (voice recognition and text-to-speech) Infrastructure:

• Resource planning via Integer Linear Programming (ILP) - with Prasad Tare

• Loadtesting tool: measured CPU utilization with <1% error

• Launched a third-party TTS system as a Google service

• Put order in the release process and the production system

• Technical Lead and Manager for a small team (3 people) for 1.5 years.

WebSearch Infrastructure: Performance tools for Mustang, the retrieval system behind Google Search. Used by 100s of engineers, for ~10 years.

Worked on the QPX airfare search engine, the engine behind Orbitz, Kayak, etc.