Brendan Duke
Senior Staff Performance Engineer at Modular
Old Toronto, Ontario, Canada
Summary
🤩
Rockstar🎓
Top SchoolBrendan Duke is a Senior Staff Performance Engineer with 11 years of experience specializing in high-performance ML inference and systems engineering, currently driving multi-node MoE and disaggregated inference performance at Modular. He has built foundational inference frameworks and kernel optimizations that enable SOTA large-model inference on both AMD and NVIDIA GPUs, and previously architected tooling and custom CUDA/C++ operators that delivered 5x performance gains in production ML pipelines. Brendan combines deep firmware and platform experience from AMD with academic rigor—he holds a PhD in Computer Science and has published at top CV conferences—allowing him to bridge low-level systems, compiler/runtime work, and applied ML. An active open-source contributor, he has contributed to PyTorch ENAS implementations, focusing on stability and hyperparameter fixes for neural architecture search. He’s also an inventor with multiple patents and a track record of shipping end-to-end ML products and scalable experiment infrastructure across cloud environments. Notably, he blends hands-on kernel and firmware debugging with distributed model engineering, making him effective at squeezing performance from both hardware and software stacks.
11 years of coding experience
8 years of employment as a software developer
Bachelor's Degree, Computer Science, Bachelor's Degree, Computer Science at McMaster UniversityDoctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at University of TorontoMasters of Applied Science, Machine Learning (School of Engineering), Masters of Applied Science, Machine Learning (School of Engineering) at University of Guelph
Github Skills (11)
pytorch10
machine-learning10
rnn-model10
n10
python10
neural-architecture-search10
algorithms8
data-structures8
algorithm8
data-structure8
tensorboard5
Programming languages (10)
Github contributions (5)
carpedm20/ENAS-pytorch
Feb 2018 - Mar 2018
PyTorch implementation of "Efficient Neural Architecture Search via Parameters Sharing"
Role in this project:
ML Engineer
ML EngineerContributions:8 commits, 2 PRs, 6 comments in 11 days
Contributions summary:Brendan primarily contributes to the implementation and debugging of a PyTorch-based implementation of Efficient Neural Architecture Search (ENAS). Their work includes fixing hyperparameter bugs related to the controller and updating function calls for compatibility with PyTorch. They also refactor code and add features for regularization, and stabilize hidden state norms, indicating a focus on improving the model's stability and performance. These changes suggest their work is directly related to the core machine learning algorithms within the repository.
pytorchparametersdeep-learningneural-architecture-searchsharing
dukebw/master-thesis-fusion
May 2019 - Sep 2020
Contributions:40 commits, 1 PR, 37 pushes in 1 year 4 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.