Xuehai Pan
Code Engineer Of Agent RL Infra at DeepSeek AI
Haidian District, Beijing, China
Summary
🤩
Rockstar🎓
Top SchoolXuehai Pan is a Ph.D. candidate at Peking University and a seasoned engineer with a decade of experience building RL/agent infrastructure and distributed AI systems. Currently leading Agent/RL infrastructure at DeepSeek AI after roles at Moonshot AI and Baichuan, he combines research rigour in multi-agent RL and game theory with production-grade engineering for large-scale RL and LLM alignment. An active open-source contributor, Xuehai has touched foundational projects—from CPython core and PyTorch compilation to testing in OpenAI Gym and build/devops work in Hugging Face Transformers—demonstrating both low-level systems skill and ML stack fluency. He also authors practical tooling like nvitop for GPU process visibility and has improved RLlib and Ray components, revealing a knack for making complex compute and infra observable and reliable. Based in Beijing, he blends academic depth with hands-on DevOps and compiler work, and often surfaces subtle correctness and performance fixes that improve robustness at scale.
10 years of coding experience
2 years of employment as a software developer
Bachelor of Science - BS, Computer Science, Bachelor of Science - BS, Computer Science at 北大Doctor of Philosophy - PhD, Computer Application Technology, Doctor of Philosophy - PhD, Computer Application Technology at Peking University
Stackoverflow
841reputation
359kreached
1answer
0questions
cuda
top-5%
Github Skills (81)
previews10
restructuredtext10
package-management10
operation10
python10
testing10
gymnasium10
bash10
refactor10
install-script10
webextension10
refactoring10
optimization10
javascript10
file-manipulation10
Programming languages (23)
Github contributions (5)
XuehaiPan/nvitop
Jan 2021 - Jan 2023
An interactive NVIDIA-GPU process viewer and beyond, the one-stop solution for GPU process management.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:10 releases, 3 reviews, 648 commits in 2 years
Contributions summary:Xuehai added a Python implementation of `nvidia-smi` using the `pynvml` library. This involved creating a Python script to display NVIDIA GPU process information, including memory usage, GPU utilization, and other metrics. The commits reflect refactoring and format updates to the initial implementation, adding features like compact mode and color support to the output. The work also included setting thresholds for memory and GPU utilization to aid in understanding system load, and improvements like tree-view and environment variable handling.
process-managementnvidia-gpumonitoringcommand-line-toolnvidia
PKU-Alignment/omnisafe
Nov 2022 - Feb 2023
OmniSafe is an infrastructural framework for accelerating SafeRL research.
Role in this project:Technical Writer
Technical WriterContributions:190 reviews, 7 commits, 23 PRs in 2 months
Contributions summary:Xuehai's commits primarily focus on updating and expanding the documentation within the OmniSafe repository. They added introductory tutorials, provided explanations of core concepts, and incorporated visual elements like cards and tab-sets to improve readability. The user also updated the documentation for specific algorithms such as FOCOPS and PCPO, and ensured that the documentation remained accurate by refactoring the packaging structure and updating the linter settings.
pytorchbenchmark-suitedeep-learningreinforcement-learningsafe-reinforcement-learning
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.