Pavel Belevich
Senior Applied Scientist, GenAI at Amazon Web Services (AWS)
New York City Metropolitan Area United States
Summary
🤩
Rockstar🎓
Top SchoolPavel Belevich is a Senior Applied Scientist specializing in GenAI with 11 years of experience building scalable ML systems and distributed infrastructure across top-tier companies including AWS, Meta, and Google. He blends deep systems and ML engineering expertise—rewriting core PyTorch random APIs, implementing TLS in Gloo, and training trillion-parameter models with record throughput—with hands-on productization work like pipeline-parallel training integrations and visual profilers. Pavel has a strong backend and distributed-systems pedigree from roles at Rubicon and PebblePost, where he designed low-latency bidding platforms and migrated large-scale data stacks to the cloud. An active open-source contributor, he’s fixed critical tracing and attention-mask issues in Hugging Face Transformers and improved PyTorch XLA support for TPUs. Based in the NYC metro area with an M.S. from MIPT, he combines rigorous applied physics/math training with practical engineering that repeatedly yields measurable performance and reliability gains.
11 years of coding experience
16 years of employment as a software developer
Master of Science Applied Physics and Mathematics, Master of Science Applied Physics and Mathematics at Moscow Institute of Physics and Technology (State University) (MIPT)
English
Stackoverflow
11reputation
5kreached
0answers
2questions
Github Skills (20)
pytorch10
c-language10
python10
machine-learning10
xla10
transformer10
cprogramming-language10
testing9
deep-learning9
nlp9
compiler8
compiler-compiler8
hub8
mapstruct6
macos6
Programming languages (8)
Github contributions (5)
huggingface/transformers
Feb 2022 - May 2022
🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
Role in this project:
ML Engineer
ML EngineerContributions:2 reviews, 6 commits, 7 PRs in 2 months
Contributions summary:Pavel contributed to the Hugging Face Transformers library by addressing issues related to the PyTorch FX tracer. They fixed bugs in the tracing process, improved compatibility with `torch.fx.Tracer`, and modified the `create_extended_attention_mask_for_decoder` method. Furthermore, the user corrected a warning message and made improvements to the models, including adding a support for a static method to the attention mask.
pythonbertspeech-recognitionstate-of-the-artflax
pytorch/xla
Jan 2020 - Apr 2020
Enabling PyTorch on XLA Devices (e.g. Google TPU)
Role in this project:Back-end Developer & QA Engineer
Back-end Developer & QA EngineerContributions:6 commits, 7 PRs, 9 comments in 2 months
Contributions summary:Pavel primarily contributed to the backend of the PyTorch XLA project, focusing on integrating PyTorch with XLA devices. Their work included renaming identifiers, modifying code related to handling generators, and modifying the test suite to address specific floating point errors. The user also updated the generator arguments, which points to their understanding of the overall structure of the framework and their contributions to the quality of the code.
pytorchxladeep-learningtpucompiler
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.