George Karpenkov
Staff Software Engineer at Tesla
Mountain View, California, United States
Summary
🤩
Rockstar🎓
Top SchoolGeorge Karpenkov is a Staff Software Engineer with 14 years of experience building high-performance ML compilers and compiler toolchains, currently working on Tesla's Dojo. He led XLA:GPU performance at Google—creating a new GPU performance team, driving Triton integration that produced 3–4x inference speedups, and institutionalizing metric-driven benchmarking and NVIDIA collaboration. His open-source contributions span TensorFlow, JAX, XLA, LLVM/Clang and Z3, with deep expertise in XLA codegen, GPU optimizations, and sanitizer/fuzzing integration for clang/libFuzzer. George combines systems-level compiler engineering with practical ML performance work, having implemented XLA improvements (GEMM, reduce-window, Triton emitters) and low-level GPU optimizations. He holds a PhD in Computer Science from Université Grenoble Alpes and has a history of shipping measurable performance gains rather than speculative research. A less obvious strength is his cross-domain impact: moving between theorem provers, compiler sanitizers, and large-scale ML stacks, he reliably turns complex low-level fixes into tangible end-user speedups.
14 years of coding experience
12 years of employment as a software developer
Doctor of Philosophy (PhD) Computer Science, Doctor of Philosophy (PhD) Computer Science at Université Grenoble AlpesThe University of Sydney
English, Russian, French
Stackoverflow
2,104reputation
216kreached
28answers
27questions
Github Skills (74)
python10
testing10
c1110
c1710
javas10
transformer-models10
code-generation10
gpu10
code-optimization10
javascript10
libfuzzer10
gpgpu10
object-oriented-programming10
ml10
verify10
Programming languages (15)
Github contributions (5)
apple/swift-clang
Aug 2017 - Feb 2019
Role in this project:
Back-end Developer
Back-end DeveloperContributions:461 commits, 4 PRs, 44 pushes in 1 year 6 months
Contributions summary:George contributed to the development of the Clang compiler, focusing on enhancements to the driver and related infrastructure. Their work included adding new functionality for fuzzing, such as the -fsanitize=fuzzer-no-link flag, and the integration of libFuzzer into the compiler-rt. They made modifications to the Darwin, Linux, and CommonArgs toolchains, as well as updates to the test suite, demonstrating expertise in compiler internals and sanitization techniques. Additional work also involved adding the support for OSObject based code, with tests that verify those aspects of the code.
openxla/xla
Mar 2019 - Jan 2023
A machine learning compiler for GPUs, CPUs, and ML accelerators
Role in this project:Back-end Developer
Back-end DeveloperContributions:321 reviews, 600 commits, 1 PR in 3 years 11 months
Contributions summary:George primarily contributes to the XLA project's low-level code generation, focusing on implementing features and optimization around core mathematical operations like matrix multiplication and reductions. Their work involves performance improvements, refactoring of indexing logic, and implementing a tree-reduction strategy. They also address bugs in the existing implementations, particularly those related to layout handling. Their contributions show a deep understanding of the underlying mechanisms, and optimization considerations related to performance.
compilercommunity-drivenmachine-learningmodular
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.