Punit Koura

Member Of Technical Staff at Microsoft AI

San Francisco, California, United States
email-iconphone-icongithub-logolinkedin-logotwitter-logostackoverflow-logofacebook-logo
Join Prog.AI to see contacts
email-iconphone-icongithub-logolinkedin-logotwitter-logostackoverflow-logofacebook-logo
Join Prog.AI to see contacts

Summary

🤩
Rockstar
🎓
Top School
Punit Koura is a machine learning engineer and software developer with 11 years of experience, currently working on vision pretraining for foundation models at Microsoft AI after contributing to Llama 2–4 at Meta. He holds an MCDS from Carnegie Mellon where he focused on ML and NLP problems like cross-lingual dependency parsing and probing contextual embeddings. Punit has a strong systems background—spanning scalable data pipelines, MapReduce, and production services at companies including Amazon, eBay, and Soroco—and applies that engineering rigor to large-scale model training and inference. He’s an active contributor to metaseq, adding document-level attention and integration work to support model-parallel transformer training. Based in San Francisco, he blends research-grade ML expertise with hands-on software engineering to move cutting-edge models from prototype into reliable infrastructure.
code11 years of coding experience
job9 years of employment as a software developer
bookBITS Pilani, Birla Institute of Technology and Science
bookMaster of Computational Data Science, Master of Computational Data Science at Carnegie Mellon University
github-logo-circle

Github Skills (9)

transformer-models10
pytorch10
machine-learning10
nlp10
huggingface10
parallelization10
python10
natural-language-processing10
testing9

Programming languages (4)

JavaRustPythonKotlin

Github contributions (5)

github-logo-circle
facebookresearch/metaseq

May 2022 - Nov 2022

Repo for external large-scale work
Role in this project:
userML Engineer
Contributions:58 reviews, 35 commits, 37 PRs in 7 months
Contributions summary:Punit primarily focused on enhancing the metaseq library, particularly in the area of model training and inference. Their work included supporting document-level attention, crucial for training model parallel models, and addressing related bugs in transformer decoder modules. The user also implemented bug fixes, code formatting changes, and added documentation. Moreover, they were involved in integration tests, ensuring compatibility with Hugging Face models.
transformersdockerscalelarge-scalehuggingface
Contributions:1 push, 1 branch in 3 years 10 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.
Request Free Trial
Punit Koura - Member Of Technical Staff at Microsoft AI