May 2020 – May 2025 · 5 yrs · New York, NY

• AI Infrastructure - Distributed ML Training for Facebook's largest-scale deep learning models, including Ads ranking, recommendation systems, and generative AI.
• Working on scalability, reliability, and efficiency of the HPC training platform that enables large-scale model training on supercomputing platforms for ML engineers, scientists, and researchers at Meta.
• Delivered end-to-end observability into components of the training stack, from data ingestion, preprocessing, trainers, checkpointing, evaluation, and more, to readily detect and debug performance, reliability, and efficiency issues in the pipeline.
• Built tooling around the Distributed ML Training Stack to detect, propagate, and root cause errors like CUDA illegal memory access, NCCL timeout issues, etc.

Oct 2018 – Mar 2020 · 1 yr 5 mos · San Francisco Bay Area

Nov 2016 – Sep 2018 · 1 yr 10 mos · Greater New York City Area

May 2014 – Jul 2016 · 2 yrs 2 mos · Gurgaon, India

• Developed iOS App Plix, solving the problem of private photo sharing. Also developed algorithm on facial recognition to determine your photos lying on your friend’s phone.
• Refined algorithm for requesting photos from contacts based on the photo’s EXIF data, GPS data, WiFi Signals and data correlation.
• Your best photos are lying on your friends' phones.
• Plix is a private photo sharing platform which lets you share entire albums from your phone as well as identifies and lets you request your friends who might have your photos.

Dec 2010 – May 2014 · 3 yrs 5 mos · India

• A self started domain to build and publish iOS Apps.