Experience

Facebook / Meta

Lead RecSys Inference Optimization (RIO) team within PyTorch AI Acceleration.
Focus on optimizing inference/serving for Meta recommendation models, supporting heterogenous hardware, and enabling MLEs to advance state-of-the-art model paradigms.

Led Config Infra (Configerator) team within Core Systems.
Developed core infrastructure for company-wide configurations, enabling fast and reliable config changes for Meta developers.

Established and led SafeChange team from the ground up.
Empowered Meta developers to make safe configuration changes quickly and easily.
Designed and implemented change safety mechanisms including automated canary and staged rollout, now part of the standard config deployment workflow at Meta.

Designed a new distributed task queuing system from scratch that processes asynchronous jobs from Yelp's backend services.
Led the reformation of the config file deployment system to allow Yelp developers to modify non-sensitive config parameters of all Yelp's services.
Modernized backend services and tools that enable developers to write efficient web and mobile API endpoints.
Pushmaster for production releases of Yelp's main codebase on bi-weekly basis for 1.5 years.

Improved tensor computation speed by 2.5x in NWChem software with a new load-balancing algorithm.
Published the results in 2013 Supercomputing Conference.