Thomas Viehmann
Principal Research Engineer at Lightning AI
Münster, North Rhine-Westphalia, Germany
Summary
🤩
Rockstar🎓
Top SchoolThomas Viehmann is a Principal Research Engineer with 26 years of experience building and optimizing machine learning systems, currently at Lightning AI and chairing the PyTorch Technical Advisory Council. He holds a PhD in Mathematics from the University of Bonn and blends deep theoretical training with hands-on PyTorch and compiler work, founding MathInf to provide ML speciality training and consulting. Thomas has contributed significantly to the TVM compiler’s ROCm backend to improve AMD GPU execution and performance—an example of his knack for bridging research and production-grade tooling. His background spans actuarial modeling and enterprise consulting at Deloitte to open-source projects like an AI-based LibreOffice translator, reflecting a rare combination of rigorous math, practical engineering, and product-aware deployment. Colleagues know him for translating complex inference and numerical challenges into dependable, high-performance implementations.
26 years of coding experience
10 years of employment as a software developer
Doctor of Philosophy (Ph.D.), Mathematics, Doctor of Philosophy (Ph.D.), Mathematics at The University of Bonn
Github Skills (11)
compiler10
roc10
compiler-compiler10
amdgpu10
deeplearning-ai9
deep-learning9
machine-learning8
python8
gpu7
cprogramming-language7
c-language7
Programming languages (11)
Github contributions (5)
apache/tvm
Jul 2019 - May 2022
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Role in this project:
ML Engineer
ML EngineerContributions:43 reviews, 44 commits, 53 PRs in 2 years 10 months
Contributions summary:Thomas primarily contributed to the ROCm (AMD GPU) backend of the TVM compiler stack, focusing on enabling and optimizing deep learning model execution on AMD GPUs. Their work included implementing ROCm-specific module loading and saving functionalities, improving the ROCm linker integration, and integrating cross-thread reductions. The user addressed performance issues by adding a workgroup size attribute to AMDGPU functions and by enabling GPU checks during compilation for ROCm. These contributions enhance TVM's support for AMD GPUs, improving usability, performance, and robustness.
metalvulkancompilertensoropencl
t-vi/AICamera
Oct 2018 - Dec 2018
Demonstration of using Caffe2 inside an Android application.
Contributions:9 commits, 2 PRs, 3 pushes in 2 months
android-applicationcaffe2android
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.