Justin Johnson
Summary
🤩
RockstarJustin Johnson is a Cofounder and research-focused software engineer with 11 years of experience building and optimizing machine learning systems, currently based in San Francisco. He blends academic rigor—PhD-level work at Stanford and a faculty role at the University of Michigan—with industry research at FAIR/Meta, contributing production-grade CUDA kernels and ML model infrastructure for projects like PyTorch3D and numerous neural-style and RNN repositories. His hands-on contributions span low-level performance engineering, Python/CUDA/torch model work, and DevOps-style build scripting, reflecting a rare combination of kernel-level optimization and end-to-end model deployment. As an entrepreneur at World Labs, he applies research insights to product-building, while his open-source edits to influential repos (e.g., PyTorch3D, densecap) show a knack for making complex algorithms practical and reproducible. An interesting detail: he has moved seamlessly between writing CUDA kernels for nearest-neighbor search and creating OSX usage-tracking tooling, demonstrating both systems-level depth and pragmatic scripting fluency.
12 years of coding experience
Github Skills (54)
content-management-system10
pytorch10
benchmark10
c-language10
python10
scripting10
image-processing10
nearest-neighbors10
testing10
machine-learning10
benchmarking10
rnn-model10
n10
appkit10
style-transfer10
Programming languages (8)
Github contributions (5)
jcjohnson/neural-style
Aug 2015 - Apr 2017
Torch implementation of neural style algorithm
Role in this project:
ML Engineer
ML EngineerContributions:134 commits, 18 PRs, 101 pushes in 1 year 7 months
Contributions summary:Justin made significant contributions to the `neural-style` repository, focused on implementing and refining the neural style transfer algorithm. Their work included adding functionality to download and utilize pre-trained VGG models, modifying hyperparameters for optimization, and incorporating features like TV regularization and multi-style support. The user also optimized memory usage and added support for OpenCL and CUDA, demonstrating a strong understanding of the underlying machine learning techniques and optimization strategies.
jcjohnson/cnn-benchmarks
Jul 2016 - Sep 2017
Benchmarks for popular CNN models
Role in this project:ML Engineer
ML EngineerContributions:52 commits, 6 PRs, 27 pushes in 1 year 2 months
Contributions summary:Justin focused on enhancing the analysis of CNN benchmark results. They developed a script to parse and generate markdown tables from the benchmark outputs, facilitating the comparison of model performance across different hardware and cuDNN versions. The user also made refinements, including fixing a bug, adjusting the table formatting, and adding support for additional GPU results (GTX 1080, CPU).
cnn-model
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.