Duncan Riach
Principal Engineer & Research Scientist at Duncan Riach
California, United States
Summary
🤩
Rockstar🎓
Top SchoolDuncan Riach is a Principal Engineer and Research Scientist based in California with eight years focused on deep learning research and large language models at NVIDIA, and a multi-decade background in computer architecture and GPU-driven systems. He has driven GPU-optimized deep learning frameworks, led efforts to make TensorFlow deterministic on GPUs (including open-source tooling and patches), and contributed reproducibility fixes to the LF AI DeepRec recommendation framework. His work blends low-level C++/CUDA kernel engineering—such as block-sparse matrix kernels and deterministic CUDA/cuDNN behaviors—with research-led experimentation on structured sparsity and performance trade-offs. Beyond engineering, Duncan has a PhD in Clinical Psychology and spent years on a sabbatical doing clinical practicum, organizational consulting, and a randomized placebo-controlled study, bringing rare cross-domain insight into human factors and team health. An inventor with multiple patents and a long history at NVIDIA dating back to its early days, he pairs rapid, high-impact engineering with leadership in complex, multi-disciplinary programs.
8 years of coding experience
8 years of employment as a software developer
BEng (hons) Electronic Engineering, BEng (hons) Electronic Engineering at University of ReadingPhD Clinical Psychology, PhD Clinical Psychology at Sofia UniversityNational Diploma Engineering, National Diploma Engineering at Brooklands Technical CollegeMS Electrical Engineering, MS Electrical Engineering at Stanford University
Github Skills (7)
machine-learning10
deep-learning10
tensorflow10
python10
cudnn9
cuda9
distributed-training8
Programming languages (4)
Github contributions (5)
DeepRec-AI/DeepRec
Jun 2019 - May 2020
DeepRec is a high-performance recommendation deep learning framework based on TensorFlow. It is hosted in incubation in LF AI & Data Foundation.
Role in this project:
ML Engineer
ML EngineerContributions:8 commits in 10 months
Contributions summary:Duncan primarily worked on implementing and testing deterministic behavior in the `bias_add` operation, critical for reproducibility in deep learning. Their commits involved modifications to kernel tests and the core `nn_ops.py` file to integrate the deterministic functionality, particularly for CUDA kernels. They also addressed deterministic behavior in the `resize_bilinear` back-propagation operation. The user also contributed to addressing multi-algorithm deterministic cuDNN convolutions.
pythondeep-learningrecommendationdistributed-trainingmachine-learning
NVIDIA/framework-reproducibility
Mar 2019 - Dec 2022
Providing reproducibility in deep learning frameworks
Contributions:8 releases, 301 commits, 6 PRs in 3 years 9 months
pytorchcaffe2deep-learninggpu-supportmachine-learning
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.