Alexei Baevski
Principal AI Researcher at Google DeepMind
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolAlexei Baevski is a Principal AI Researcher based in San Francisco with a decade of experience bridging deep research and production-grade ML systems. He has held senior research roles at Meta and Character.AI and now leads advanced AI work at Google DeepMind, specializing in generative models, speech and self-supervised learning, sparse architectures, and long-context reasoning. A key contributor and maintainer of the widely used fairseq toolkit, he has improved generation efficiency and batching logic—work that directly impacts large-scale sequence modeling pipelines. His publications and invited talks reflect a sustained influence in SSL and speech (wav2vec/data2vec), while his background in software engineering and systems leadership informs a pragmatic approach to deploying research. Unexpectedly, his track spans both low-latency production search features and cutting-edge model research, giving him rare fluency across product, infrastructure, and foundational ML.
10 years of coding experience
17 years of employment as a software developer
Bachelor of Science (BSc) Computer Science - Software engineering specialization, Bachelor of Science (BSc) Computer Science - Software engineering specialization at University of Toronto
English, Russian
Github Skills (10)
transformer10
machine-learning10
pytorch10
artificial-intelligence10
backend10
python10
algorithms8
data-structures8
algorithm8
data-structure8
Programming languages (8)
Github contributions (5)
facebookresearch/fairseq
Apr 2018 - Dec 2022
Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:24 reviews, 561 commits, 14 PRs in 4 years 9 months
Contributions summary:Alexei focused on improving the interactive mode of the fairseq library by enhancing alignment printing and fixing optimization history. They addressed comments to improve the trainer.py module and implemented caching for the multihead attention module, optimizing the processing of last time steps during generation. Further, the user worked on removing completed sentences from the batching process, enabling support for uneven lengths in batches.
pytorchnlpsequencepythontransformer-architecture
alexeib/Jalex.Common
Dec 2015 - Jan 2017
Contributions:55 pushes, 1 branch in 1 year 1 month
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.