Roman Grundkiewicz
Principal Researcher at Microsoft
Redmond, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolRoman Grundkiewicz is a Principal Researcher at Microsoft with 14 years of experience specializing in neural machine translation, NLP, and high-performance C++/CUDA implementations. He combines deep academic roots—a PhD in Computer Science and years of research and teaching—with industry impact across Microsoft, Samsung R&D, WIPO, and the University of Edinburgh. Roman’s work spans algorithmic research (automatic grammatical error correction, domain adaptation, low-resource MT) and production-focused optimization, including GPU-accelerated tensor operations for the well-known marian-nmt project. He has a track record of making translation systems both more accurate and efficient, often bridging research prototypes and deployable systems in patent and real-world domains. Based in Redmond, he pairs rigorous experimentation with hands-on systems engineering, frequently contributing low-level performance improvements that are easy to overlook but crucial in practice.
14 years of coding experience
14 years of employment as a software developer
Doctor of Philosophy Computer Science, Doctor of Philosophy Computer Science at Uniwersytet im. Adama Mickiewicza w Poznaniu
English, Polish
Github Skills (16)
tensorrt10
cuda10
kernel10
gpu-programming10
c-language10
tensor10
tensorflow10
cprogramming-language10
operation10
linear-algebra9
neural-machine-translation9
algorithms8
algorithm8
numeric8
numerical-methods8
Programming languages (12)
Github contributions (5)
marian-nmt/marian
Sep 2016 - Jan 2022
Fast Neural Machine Translation in C++
Role in this project:
Back-end Developer
Back-end DeveloperContributions:993 commits, 2 PRs, 11 pushes in 5 years 5 months
Contributions summary:Roman's contributions primarily involve modifications to the `src/kernels/tensor_operators.cu` file within the "marian-nmt/marian" repository, which is a C++-based fast neural machine translation framework. The code changes suggest the implementation of new functionalities for tensor operations and modifications related to gradient calculation and matrix manipulation. Their work involved the use of CUDA for GPU acceleration, demonstrating a focus on optimizing tensor operations for performance within the neural machine translation context.
machine-translationtranslationneural-machine-translationcudagpu
marian-nmt/marian-benchmarks
Dec 2017 - Dec 2019
Performance benchmarks for Marian
Contributions:1 review, 38 commits, 10 pushes in 2 years
benchmarkingperformance-benchmarksbenchmarkperformancebenchmarks
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.