Brennan Saeta
Member Of Technical Staff, Manager at Anthropic
Kirkland, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolBrennan Saeta is a seasoned software engineering leader with 14 years of experience building large-scale ML infrastructure and accelerator-aware systems, currently managing teams at Anthropic focused on training and inference across GPUs, TPUs, and Trainium. Previously at Google Brain he co-founded the Cloud TPU project, led the JAX Engagements Team that helped power PaLM and Gemini, and shipped core JAX features like pjit and Global Device Array. He combines low-level performance tuning and distributed systems design with product-facing delivery, from micro-optimizations to co-designing hardware and models. An active open-source contributor, Brennan worked on Swift for TensorFlow and experiments in fastai, and helped productionize continuous batching and other inference optimizations now widely referenced in LLM deployments. He holds an MS and BS in Computer Science from Stanford and is known for turning research-grade code into robust, production systems.
14 years of coding experience
13 years of employment as a software developer
Master of Science (MS) Computer Science, Master of Science (MS) Computer Science at Stanford University
French
Github Skills (24)
python10
optimizers10
model-driven10
machine-learning10
optimizer10
tpu10
model-building10
callback10
deeplearning-ai10
deep-learning10
tensorflow10
swift10
differentiable-programming10
modeling10
model-driven-development10
Programming languages (12)
Github contributions (5)
tensorflow/swift-apis
Feb 2019 - Sep 2020
Swift for TensorFlow Deep Learning Library
Role in this project:
Back-end Developer
Back-end DeveloperContributions:4 reviews, 41 commits, 103 PRs in 1 year 6 months
Contributions summary:Brennan focused on enhancing the `swift-apis` repository, which supports Swift for TensorFlow. They contributed by adding default values to layer initializers, improving optimizer usability, and making learning rates dynamically settable. Their work included refactoring and improving error messages, along with addressing issues in batch gathering. Furthermore, they removed deprecated datasets and addressed deprecation warnings.
differentiable-programmingswift-for-tensorflowdeep-learningmachine-learningdeep-learning-library
tensorflow/tpu
Aug 2017 - Mar 2019
Reference models and tools for Cloud TPUs.
Role in this project:Back-end Developer
Back-end DeveloperContributions:53 commits, 54 PRs, 36 pushes in 1 year 7 months
Contributions summary:Brennan's commits focus on integrating internal changes into the public repository for the `tensorflow/tpu` project. These changes include modifications to Python files related to LSTM models on the PTB dataset and adjustments to Magenta sequence example libraries, suggesting enhancements to machine learning model training and data handling. Additionally, there are alterations to Go tools related to CTPU commands and configuration, implying contributions to the infrastructure or tooling used for managing TPUs.
cloud
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.