Qiumin Xu
Software Engineering Manager at Google
California, United States
Summary
🤩
Rockstar🎓
Top SchoolQiumin Xu is a software engineering manager with 13 years of experience leading high-performance ML and systems teams, currently directing the AI Experience team for Google Tensor to deploy and optimize generative, speech, and imaging models. With a Ph.D. in electrical engineering from USC and deep roots in GPU architecture, she has driven tooling and performance work across TensorFlow, TensorBoard, and TPU stacks as a tech lead and staff engineer. An active open-source contributor, she upstreamed and hardened JAX features like named_call and improved TensorBoard profiling UI and memory views, helping bridge model development with production profiling and conversion workflows. Comfortable operating at the intersection of research and production, she combines hardware-aware performance insight with practical deployment experience to accelerate real-world AI applications.
13 years of coding experience
12 years of employment as a software developer
Doctor of Philosophy (Ph.D.) Electrical Engineering, Doctor of Philosophy (Ph.D.) Electrical Engineering at University of Southern CaliforniaBachelor of Science (B.S.) EECS, Bachelor of Science (B.S.) EECS at Peking University
Chinese, English, Japanese
Stackoverflow
11reputation
232reached
1answer
0questions
Github Skills (31)
javascript10
performance-analytics10
performance-monitor10
python10
code-profiling10
machine-learning10
ui-design10
uid10
performance-analysis10
tensorflow10
front-end-development10
profiling10
html10
web-development10
jax10
Programming languages (4)
Github contributions (5)
tensorflow/tensorboard
Oct 2017 - Mar 2020
TensorFlow's Visualization Toolkit
Role in this project:
Full-stack Developer
Full-stack DeveloperContributions:41 commits, 61 PRs, 13 pushes in 2 years 4 months
Contributions summary:Qiumin primarily contributed to the TensorBoard profiling plugin, focusing on the user interface and performance enhancements. They fixed bugs within the `tf-op-table.html` and `tf-op-details.html` files, related to the op profile tool, and its display of data. They also implemented features such as the slider and toggle button to show top K time consuming ops per category and added documentation to the profile plugin. Further work included the addition of a memory viewer and line charts to the tool, and integration for the profile plugin to take per host profile traces.
machine-learningtensorflowvisualization
jax-ml/jax
Dec 2020 - Dec 2020
Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
Role in this project:ML Engineer
ML EngineerContributions:16 reviews, 3 commits, 7 PRs in 5 days
Contributions summary:Qiumin primarily focused on integrating and extending the JAX library, specifically implementing a `named_call` feature. They added the implementation, test cases, and public API for `named_call`, allowing users to name subcomputations in JAX for better debugging and integration with tools like the TensorFlow Profiler. The user also moved the `named_call` implementation between different files, refactoring it for better structure. Additionally, they contributed an example of converting a Flax model for MNIST to TensorFlow Lite, including quantization.
pytorchpythonjitautomatic-differentiationgpu
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.