Worked on various areas: Image/Video Generation training/inference Infra/Tooling, GenAi Model Deployment and API Serving, GenAI Batch Inference

Proposed and co-authored an industry-leading video decoding library - TorchCodec - beating TorchVision/Decord on performance (accuracy and speed). Within Meta, it is used by MovieGen and Llama multi-modality. The work was presented in Pytorch Conference 2024 and later open-sourced.

Designed and implemented an efficient video clip caching mechanism that improved the training speed of video understanding models by 5–10×, used by content understanding models across reels, integrity and ads.

Led the deprecation of Facebook’s Face Recognition system across the Family of Apps represented one of the largest shifts in facial recognition usage in technology history, affecting billions of users globally and setting new standards for privacy and responsible AI.

Tech lead for Facebook OCR, a company-wide optical character recognition service used across products for content understanding, accessibility, and integrity.

Key contributor to device-based OTA infra for Oculus Devices (Meta Quests, Meta Ray-ban), built release and firmware OTA infra and tooling.

Worked at Amazon Chime team on a cloud-based telephony infrastructure project.