Jason Furmanek
Principal Engineer at IBM
Austin, Texas, United States
Summary
🤩
RockstarJason Furmanek is a Principal Engineer in Austin with 11 years of experience specializing in GPU compiler backends and ROCm support for machine learning stacks. At AMD and previously leading TensorFlow work at IBM, he bridges low-level compiler engineering and DevOps, shipping compatibility and performance fixes across Triton, XLA, and TensorFlow ROCm ports. His open-source contributions include enabling ROCm6/ROCm5 compatibility, improving CI for conda-forge, and unlocking AMD backend features that directly impact PyTorch and XLA performance. Known for pragmatic refactors—like removing legacy type converters and mapping precise math to LLVM—he combines systems-level rigor with a developer-focused approach to build and test infrastructure.
11 years of coding experience
Github Skills (21)
docker10
c-language10
continuous-integration10
python10
travis10
compiler-development10
llvm10
gpu-programming10
dockers10
cicd10
roc10
triton10
tensorflow10
build-automation10
xla10
Programming languages (14)
Github contributions (5)
ROCm/tensorflow-upstream
Nov 2021 - Jan 2023
TensorFlow ROCm port
Role in this project:
DevOps Engineer
DevOps EngineerContributions:253 reviews, 438 commits, 1169 PRs in 1 year 2 months
Contributions summary:Jason's commits primarily focus on adapting and maintaining the build and test infrastructure for ROCm support within the TensorFlow project. They updated build scripts to use ROCm 5.4.0, adjusted includes for roctracer, and modified test scripts by increasing timeout values, demonstrating a focus on ensuring the build and testing environment's stability and compatibility with different ROCm versions. Additionally, the user fixed setup scripts and updated build tools, indicating their involvement in refining the build process.
tensorflow
conda-forge/conda-smithy
Aug 2018 - Aug 2019
The tool for managing conda-forge feedstocks.
Role in this project:DevOps Engineer
DevOps EngineerContributions:10 commits, 1 PR, 17 comments in 1 year
Contributions summary:Jason primarily contributed to improving the continuous integration setup for the conda-smithy project. Their work included enabling Travis builds for the ppc64le architecture and adjusting tests to accommodate new build variants. They also enabled yum_requirements for Travis and added related news documentation. The user's contributions streamlined build processes.
conda-forgecontinuous-integration
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.