Dominik Grewe
Principal Software Engineer at Isomorphic Labs
London, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolDominik Grewe is a Principal Software Engineer based in London with 11 years of experience building high-performance software and research-facing tooling for AI. He spent over a decade at DeepMind after a PhD from the University of Edinburgh focused on mapping parallel programs to heterogeneous GPU systems, and now applies that expertise at Isomorphic Labs. Dominik has deep low-level CUDA and backend experience—contributing notable optimizations to the Torch ecosystem (cutorch, cunn, torch7) including new modules, efficient batch matrix ops, and RNG and tensor primitives. He combines research rigor with production instincts, having worked on compiler, benchmarking, and performance teams at Google, NVIDIA and ARM. Colleagues rely on him for solving hard performance bottlenecks and for clean, well-tested implementations that bridge academia and industry. A less obvious strength is his long history of teaching and mentoring, which helps him translate complex systems ideas into practical team knowledge.
11 years of coding experience
12 years of employment as a software developer
PhD Informatics, PhD Informatics at The University of EdinburghBSc Computer Science, BSc Computer Science at Universität des Saarlandes
English, German, French
Stackoverflow
21reputation
290reached
2answers
0questions
Github Skills (29)
thrust10
pytorch10
c-language10
matrix10
operation10
tensorrt10
matrix-multiplication10
pytorch-lightning10
c1110
c1710
lua10
mat10
deep-learning10
tensorflow10
cuda10
Programming languages (8)
Github contributions (5)
torch/cutorch
Aug 2014 - Feb 2016
A CUDA backend for Torch7
Role in this project:
Back-end Developer
Back-end DeveloperContributions:63 commits, 54 PRs, 11 pushes in 1 year 5 months
Contributions summary:Dominik focused on extending the CUDA backend for Torch7, specifically addressing the implementation of missing functionalities. They added support for the three-argument version of the `cdiv` operation and implemented the `cpow` and `tpow` operations. Additionally, the user worked on improving the random number generator and implemented cumsum and cumprod operations. Their contributions also involved refactoring the code to better utilize the THCudaState.
cudagpubackendcuda-backend
torch/torch7
Jan 2015 - Feb 2016
http://torch.ch
Role in this project:Backend Developer
Backend DeveloperContributions:25 commits, 25 PRs, 9 pushes in 1 year
Contributions summary:Dominik primarily focused on implementing and optimizing core functionalities within the Torch7 library. They added batch matrix-matrix multiplication (bmm) functionality, rewrote bmm to utilize baddbmm for improved efficiency, and introduced gather and scatter operations. Further contributions include making the totable and permute functions available as torch.* functions and other adjustments such as passing an optional allocator to storage and various bug fixes.
libtorchtorchc-plus-plustensorautograd
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.