Avital Oliver
Neural Network Plumber at The Global Math Circle
San Francisco, California, United States
Summary
🤩
RockstarAvital Oliver is a software engineer with 20 years of experience, currently a Neural Network Plumber at OpenAI in San Francisco, focused on ergonomic neural network libraries and memory- and compute-efficient scaling of massive vision models. As tech lead on Google Flax, Avital helped turn the JAX library into the de facto standard for large models—contributing key features like weight/group normalization and robust RNG/dropout handling used in MLPerf-winning training runs and products across Google and Hugging Face. Their career uniquely spans embedded device programming, realtime full‑stack systems and cutting‑edge ML research, from a 3D renderer in x86 assembly to a Google Wave synchronization robot. They also lead the Global Math Circle and publish work on scaling and efficient architectures, blending deep implementation fluency with intentional API and library design.
17 years of coding experience
14 years of employment as a software developer
English, Hebrew
Stackoverflow
1,542reputation
46kreached
15answers
0questions
meteor
top-5%
Github Skills (35)
operational-transformation10
javascript10
python10
back-end-development10
testing10
machine-learning10
flax10
javascripts10
neural-networks10
operational-transform10
transform10
back-end10
jax10
backend10
meteor10
Programming languages (8)
Github contributions (5)
google/flax
Jan 2020 - Feb 2022
Flax is a neural network library for JAX that is designed for flexibility.
Role in this project:
ML Engineer
ML EngineerContributions:2 releases, 337 reviews, 238 commits in 2 years 1 month
Contributions summary:Avital's contributions primarily focused on the development and improvement of machine learning components within the Flax framework. Their work included modifying the dropout implementation, fixing issues related to random number generation (RNG) streams in dropout, and refining examples, such as the LM1B example pipeline. They introduced new features like adding a dtype argument to `DenseGeneral`, weight normalization, and group normalization layers, along with improving the Module class methods. The user also added a momentum contrast for unsupervised visual representation learning.
deep-learningneural-networksneural-networkflaxjax
share/sharedb
Mar 2016 - Sep 2016
Realtime database backend based on Operational Transformation (OT)
Role in this project:Back-end Developer
Back-end DeveloperContributions:73 commits, 23 PRs, 46 pushes in 6 months
Contributions summary:Avital primarily focused on improving the stability and maintainability of the ShareDB backend. They initialized object properties for optimization, refactored tests to be more robust and less flaky, and implemented error handling with stack traces. Additionally, the user made modifications to the database interactions and test setup, demonstrating a good understanding of the backend architecture. The contributions improved code quality and testing.
backend-databaserealtimetransformationreal-timeoperational
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.