Bert Maher
Member Of Technical Staff at Anthropic
United States
Summary
🤩
Rockstar🎓
Top SchoolBert Maher is a compiler-focused software engineer with 13+ years building and optimizing compilers and performance tooling across domains from ML hardware (Glow, PyTorch/TorchScript, functorch, Triton) to mobile bytecode (ReDex) and binary translation (Transmeta/Intel MoonRun). He combines deep systems and computer-architecture knowledge with pragmatic profiling skills to squeeze cycles and reduce latency in real-world stacks, and has led compiler efforts at Facebook/Meta before joining Anthropic to focus on inference and performance. His open-source footprint includes substantive backend work on high-profile projects like PyTorch/functorch and Triton, adding eager compilation support and improving module compilation and caching. Bert’s contributions span both low-level memory- and layout-sensitive optimizations (HFSort, ProguardMap, Dex handling) and developer-facing build systems (Buck, PyTorch builder with LLVM integration). He holds advanced computer-science training from UT Austin (PhD/MS) and a BS from Yale, pairing research depth with long-term production impact. An underappreciated strength is his consistent ability to simplify configurations and tests while improving runtime efficiency, reflecting a taste for maintainable high-performance engineering.
13 years of coding experience
15 years of employment as a software developer
Ph.D. Computer Science, Ph.D. Computer Science at The University of Texas at AustinB.S. Electrical Engineering and Computer Science, B.S. Electrical Engineering and Computer Science at Yale University
Github Skills (39)
code-optimization10
pytorch10
c-language10
back-end-development10
python10
llvm10
testing10
machine-learning10
bash10
cicd10
java10
gradient10
javas10
compiler-design10
tensorflow10
Programming languages (10)
Github contributions (5)
pytorch/glow
Feb 2018 - Feb 2020
Compiler for Neural Network hardware accelerators
Role in this project:
Back-end Developer
Back-end DeveloperContributions:455 commits, 512 PRs, 208 pushes in 2 years
Contributions summary:Bert's commits focus on removing configuration options and simplifying code related to training configurations within the Glow framework. They removed parameters and made updates to incorporate these code changes in tests such as "tests/unittests/InterpreterTest.cpp". These modifications appear to be aimed at improving the efficiency and maintainability of the training setup within the compiler for neural network hardware accelerators.
hardware-acceleratorscompilerneural-networkacceleratorshardware
facebook/redex
Mar 2016 - Sep 2017
A bytecode optimizer for Android apps
Role in this project:Back-end Developer
Back-end DeveloperContributions:2 releases, 189 commits, 4 PRs in 1 year 6 months
Contributions summary:Bert primarily contributed to the optimization of the Redex bytecode optimizer for Android applications. Their work involved fixing bugs related to argument handling, and memory safety concerns. Additionally, they improved the ProguardMap, and added code to manage and manipulate Dex files, including the handling of String and Type elements. The user also refactored the configuration handling process.
android-appsandroidbytecodekotlinoptimizer
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.