Nicolas Koumchatzky
Compute Platform Lead
New York, New York, United States
Summary
👤
Senior🎓
Top SchoolNicolas Koumchatzky is a compute platform leader with a decade of experience architecting and scaling ML infrastructure for high-stakes environments, currently guiding Hudson River Trading's AI and algorithm research platform. He previously built and led NVIDIA's AI infrastructure across multiple locations, directing MagLev—the production-grade ML platform—and spearheading a GPU-based Recommender Systems initiative with a 200-person team. At Twitter, he led Cortex, overseeing the core ML platform, shaping strategy, team health, and high-impact ML workflows, while mentoring a dispersed engineering organization. An active contributor to open-source ML tooling, he enhanced the Lua Torch autograd ecosystem (torch-autograd) with automatic differentiation capabilities and modular, clonable components. His academic foundation blends statistics and financial mathematics with advanced science/engineering, complemented by prior research and teaching roles, giving him a strong blend of quantitative rigor and production-grade pragmatism.
10 years of coding experience
8 years of employment as a software developer
Ecole polytechnique
English, French
Github Skills (7)
neural-network10
lua10
pytorch10
automatic-differentiation10
autograd10
machine-learning9
testing8
Programming languages (3)
Github contributions (5)
twitter-archive/torch-autograd
Oct 2015 - Jan 2016
Autograd automatically differentiates native Torch code
Role in this project:
ML Engineer
ML EngineerContributions:8 commits, 6 PRs, 20 pushes in 2 months
Contributions summary:Nicolas contributed significantly to the `torch-autograd` repository by adding functionality for automatic differentiation in Torch. Their work includes enabling the `torch.index`, `torch.narrow`, and `torch.select` functions within the autograd system, along with associated testing. They introduced `AutoModule` and `AutoCriterion` classes, allowing for the automated construction and differentiation of custom neural network modules and loss functions defined using standard Lua functions. Further contributions include making these auto-generated modules clonable, savable and loadable.
pytorchautomatic-differentiationtorchmachine-learningautodiff
nkoumchatzky/torch7
Jul 2015 - Feb 2016
Contributions:2 PRs, 8 pushes, 2 branches in 7 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.