Andrew Tulloch
AI Researcher at Meta
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolAndrew Tulloch is an AI researcher and systems engineer with 14 years of experience building and optimizing ML infrastructure across industry-leading labs, including Meta and OpenAI, and now based in San Francisco. He combines deep theory (Cambridge MMath/Statistics background) with hands-on systems work—contributing to high-profile open-source projects such as PyTorch-related repositories, TVM, and FBGEMM—focusing on CUDA kernels, quantized kernels, and compiler/tooling performance. At Meta he rose to Distinguished Engineer driving PyTorch and ML systems, and at OpenAI he worked on GPT-4 pretraining and o-series reasoning, blending model-scale research with production-focused engineering. His contributions show a pattern of improving low-level performance and reliability (memory leaks, thread safety, DWARF/LLVM verifier, ARM_NEON fixes), revealing a penchant for squeezing hardware efficiency out of ML stacks. Colleagues know him for translating complex numerical ideas into robust, well-tested code that scales from CUDA kernels to compiler-level improvements.
14 years of coding experience
13 years of employment as a software developer
Master of Mathematical Statistics Statistics Machine Learning, Master of Mathematical Statistics Statistics Machine Learning at University of CambridgeThe University of SydneyChrist Church Grammar School, Claremont
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (36)
unit-testing10
pytorch10
caffe10
c-language10
fba10
emm10
llvm10
gpu-programming10
machine-learning10
kernel10
lua10
deep-learning10
f10
performance-optimization10
ai10
Programming languages (16)
Github contributions (5)
facebookarchive/caffe2
Nov 2016 - Feb 2018
Caffe2 is a lightweight, modular, and scalable deep learning framework.
Role in this project:
ML Engineer
ML EngineerContributions:130 commits, 1 PR, 6 comments in 1 year 2 months
Contributions summary:Andrew made several contributions related to the Caffe2 deep learning framework. These involved fixing issues related to memory management, thread safety, and performance optimization within core operators and related test code. The user also implemented a new operator and addressed issues in the ARM_NEON codepaths, demonstrating a focus on improving the framework's functionality and efficiency. They further improved the performance of instance normalization for NCHW format.
pytorchscalablecaffe2deep-learningml
facebookarchive/fblualib
Oct 2014 - Jul 2015
Facebook libraries and utilities for Lua
Role in this project:Back-end Developer
Back-end DeveloperContributions:12 commits in 8 months
Contributions summary:Andrew primarily contributed to bug fixes and enhancements within the Facebook Lua library. Their work involved addressing lint warnings, fixing unused variables, and resolving issues in the LuaUnit testing framework. The user also implemented a `defaultdict` functionality and debugged the fbpython bridge. They demonstrated proficiency in Lua and familiarity with testing methodologies and code optimization.
luafacebook
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.