Tamás Fehér
AI Developer Technology Manager at NVIDIA
Neufahrn bei Freising, Bavaria, Germany
Summary
🤩
Rockstar🎓
Top SchoolTamás Fehér is an AI developer technology manager at NVIDIA with a decade-plus track record of extracting peak performance from GPUs for ML, HPC and scientific codes. He combines a PhD in plasma physics with hands-on expertise in CUDA, MPI/OpenMP, C/C++, Fortran and Python to optimize deep learning training/inference, implement new parallel algorithms, and port large scientific simulations to accelerators. At NVIDIA he led teams commoditizing vector search and high-performance ML primitives, and his open-source contributions include performance work in high-impact projects like TensorFlow-TensorRT and RAPIDS (raft, cuML). Known for improving both numerical stability (e.g., SVM SMO solver) and practical tooling (dynamic-shape TensorRT examples, IVF index serialization), he bridges research-grade modeling and production-grade engineering. Based in Bavaria, he still draws on plasma-physics modeling intuition when designing algorithms, giving him a rare mix of domain science and systems-level optimization.
7 years of coding experience
13 years of employment as a software developer
Doctor of Philosophy (Ph.D.) Plasma Physics, Doctor of Philosophy (Ph.D.) Plasma Physics at Universität GreifswaldMSc (with Distinction) Physics and Informatics, MSc (with Distinction) Physics and Informatics at Eötvös Loránd UniversityLicentiate of Engineering Nuclear Engineering Applied Physics, Licentiate of Engineering Nuclear Engineering Applied Physics at Chalmers University of Technology
English, German, Hungarian
Stackoverflow
11reputation
3kreached
1answer
0questions
Github Skills (28)
algorithm10
code-optimization10
algorithms10
c-language10
python10
tensorrt10
gpu-programming10
machine-learning10
machine-learning-algorithms10
tensorflow10
optimisation10
cuda10
numerical-optimization10
cprogramming-language10
linear-algebra10
Programming languages (5)
Github contributions (5)
rapidsai/raft
Nov 2020 - Jan 2023
RAFT contains fundamental widely-used algorithms and primitives for machine learning and information retrieval. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications.
Role in this project:
Back-end Developer / ML Engineer
Back-end Developer / ML EngineerContributions:584 reviews, 16 commits, 54 PRs in 2 years 2 months
Contributions summary:Tamás primarily contributed to the `raft` library by implementing and refining linear algebra primitives with CUDA acceleration. The commits showcase the implementation of a Cholesky rank one update, incorporating error checking and a regularization parameter, demonstrating expertise in numerical methods and GPU programming. Further contributions included refactoring and enhancing the `map_then_reduce` operation and also added the serialization and deserialization features for IVF-PQ and IVF Flat indexes, demonstrating the ability to work with data structures and storage. The user’s work also shows involvement with optimizing algorithms related to nearest neighbor search within the repository's machine learning and information retrieval focus.
sciencedata-sciencemachine-learninggraph-data-sciencefundamental
rapidsai/cuml
Feb 2019 - Jul 2021
cuML - RAPIDS Machine Learning Library
Role in this project:Back-end Developer
Back-end DeveloperContributions:142 reviews, 300 commits, 45 PRs in 2 years 5 months
Contributions summary:Tamás primarily worked on implementing and refining the SMO (Sequential Minimal Optimization) algorithm for Support Vector Machine (SVM) training, a core component of the cuML library. Their contributions involve refactoring code, moving working set selection to a separate class, and optimizing the kernel cache, showcasing expertise in machine learning algorithms and their implementation. Additionally, the user added tests for the SMO block solver and further improved the numerical stability of the solver. These changes indicate an effort to enhance the performance and reliability of the SVM implementation.
cudacumlnvidiadata-sciencegpu
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.