Jacob Kahn
AI Research TLM, FAIR, MSL at University of Pennsylvania
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolJacob Kahn is an AI researcher and systems engineer with 11 years of experience, currently leading FAIR code generation and reasoning efforts at Meta AI while also teaching GPUs & ML Systems as CS faculty at UPenn. He specializes in model-system co-design—optimizing architecture, computation, compilation, and communication for large-scale training—and has applied that expertise across multimodal, retrieval, and speech projects. His background blends production-grade C++/GPU backend work (notably contributions to libraries like Flashlight and ArrayFire and porting efforts in vcpkg) with research on code generation and reasoning. Jacob’s cross-disciplinary training in computer science and economics (Wharton + Penn engineering) helps him bridge algorithmic depth with practical system performance. He is known for improving low-level performance plumbing—MKL-DNN CPU kernels, backend loaders, and memory manager frameworks—that materially speed model training. Based in San Francisco, he brings both academic rigor and hands-on engineering to scale cutting-edge AI systems.
11 years of coding experience
1 year of employment as a software developer
Bachelor of Science (B.S.), Economics | Concentrations in Statistics & Operations Research and Management, Bachelor of Science (B.S.), Economics | Concentrations in Statistics & Operations Research and Management at The Wharton SchoolJerome Fisher M&T ProgramMaster of Science in Engineering (M.S.E.), Computer Science, Master of Science in Engineering (M.S.E.), Computer Science at University of Pennsylvania
English, Spanish, Chinese, Vietnamese
Github Skills (27)
performance-monitor10
c-language10
lib10
edn10
operation10
back-end-development10
package-management10
tensorrt10
machine-learning10
cmake10
c1110
c1710
performance-analysis10
tensorflow10
gpu10
Programming languages (8)
Github contributions (5)
flashlight/flashlight
Dec 2018 - Jan 2023
A C++ standalone library for machine learning
Role in this project:
ML Engineer
ML EngineerContributions:3 releases, 102 reviews, 447 commits in 4 years 1 month
Contributions summary:Jacob primarily contributed to the optimization and implementation of machine-learning functionalities within the `flashlight/flashlight` repository, a library dedicated to machine learning in C++. Their work includes fixing typos in build documentation, modifying include paths for flashlight assets, and implementing a Pool2D for the CPU backend using MKL-DNN. They also relaxed cuDNN version requirements and made enhancements to the tensor operations and the framework, demonstrating a focus on performance and usability. The commits show an engagement with enhancing the building tools and core mathematical functions to enable the performance of other components.
cppdeep-learningstandalone-libraryc-plus-plusmachine-learning
arrayfire/arrayfire
Oct 2018 - Apr 2022
ArrayFire: a general purpose GPU library.
Role in this project:Backend Engineer
Backend EngineerContributions:1 review, 12 commits, 9 PRs in 3 years 6 months
Contributions summary:Jacob focused on improving the ArrayFire library's functionality and performance. Their contributions include adding a custom path for loading dynamic backend libraries, fixing BLAS gemm function generators, and properly including spdlog for macro usage. Furthermore, they worked on adding a framework for extensible memory managers and optimizing the JIT for sequential casts. They also contributed a Conanfile for the Linux binary installer.
cudaarrayfirecppgpuscientific-computing
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.