Katherine Lee
Researcher at OpenAI
San Francisco, California, United States
Summary
👤
Senior🎓
Top SchoolKatherine Lee is a research engineer with 11 years of experience specializing in privacy and security for large language models, currently a Researcher at OpenAI after research roles at Google Research/Brain and DeepMind. She has a strong engineering background in scalable model training and data pipelines, contributing to high-profile open-source projects like Mesh TensorFlow, Tensor2Tensor, and the Text-to-Text Transfer Transformer. Her work bridges hands-on ML engineering—adding continuous evaluation, deterministic data handling, and novel training features—with research-driven investigation into model vulnerabilities. Based in San Francisco and trained in Operations Research and Financial Engineering at Princeton, she combines rigorous quantitative thinking with practical system-building. An often-overlooked strength is her focus on evaluation infrastructure and reproducible preprocessing, which quietly improves robustness across large-model experiments.
12 years of coding experience
7 years of employment as a software developer
Optics and Modern Physics Research Lab, Optics and Modern Physics Research Lab at Thomas Jefferson High School for Science and TechnologyOperations Research and Financial Engineering, Operations Research and Financial Engineering at Princeton University
English, Chinese
Github Skills (18)
python10
evaluation10
machine-learning10
tpu10
generative-adversarial-network10
eval10
deep-learning10
trainings10
tensorflow10
natural-language-processing10
nlp10
modeling10
image-processing9
data-pipelines9
data-preprocessing9
Programming languages (3)
Github contributions (5)
tensorflow/mesh
Apr 2019 - Jun 2021
Mesh TensorFlow: Model Parallelism Made Easier
Role in this project:
ML Engineer
ML EngineerContributions:28 commits in 2 years 2 months
Contributions summary:Katherine made several contributions focused on enhancing the model training and evaluation pipeline within the Mesh TensorFlow framework. These contributions include the addition of continuous evaluation capabilities, allowing for real-time assessment of model performance, and the specification of evaluation metrics based on components. Further refinements involved refactoring the decoding process and refactoring and expanding the documentation.
tensorflow
tensorflow/tensor2tensor
Aug 2017 - Sep 2019
Library of deep learning models and datasets designed to make deep learning more accessible and accelerate ML research.
Role in this project:ML Engineer
ML EngineerContributions:9 commits in 2 years
Contributions summary:Katherine made several contributions focused on enhancing the capabilities of the Tensor2Tensor library. These include adding new features like "strokes SpaceID" and implementing Gaussian label smoothing. Further contributions include the addition of image summary metrics and the integration of a vanilla GAN model. Additionally, the user made adjustments, such as piping the decode_reference flag, to improve the functionality and configuration options within the system.
deep-learningmachine-learningmachine-translationreinforcement-learningtpu
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.