Xiaoย Wang
Applied Scientist at Microsoft
Redmond, Washington, United States
Summary
๐คฉ
Rockstar๐
Top SchoolXiao Wang is an Applied Scientist and MS candidate in ECE at UIUC with eight years of engineering and research experience focused on LLMs, RAG, and applied ML systems. At Microsoft, Xiao built a multilingual LLM code-translation and unit-test automation framework that boosted test coverage by 42% and helped create a multilingual code-completion benchmark for GitHub Copilot across 10 languages. Their research on iterative and verifiable RAG for medical QA achieved SOTA results and led to publications and collaborations with Microsoft and NIH, demonstrating a knack for translating academic advances into practical, auditable systems. An active open-source contributor, Xiao optimized LLM reinforcement-learning tooling (verl) with sequence parallelism and SFT performance improvements used in production-grade workflows. Based in Redmond, they seek new graduate roles in SDE, MLE, or Applied Science, blending deep research chops with hands-on system engineering.
7 years of coding experience
1 year of employment as a software developer
Master of Science - MS, ECE, Master of Science - MS, ECE at University of Illinois Urbana-ChampaignBachelor of Engineering - BE, Electrical Engineering, Bachelor of Engineering - BE, Electrical Engineering at Zhejiang University
Github Skills (7)
transformers10
pytorch10
machine-learning10
large-language-models10
python10
llm10
reinforcement-learning9
Programming languages (7)
Github contributions (5)
volcengine/verl
Nov 2024 - Mar 2025
verl: Volcano Engine Reinforcement Learning for LLMs
Role in this project:
ML Engineer
ML EngineerContributions:11 reviews, 7 PRs, 23 comments in 5 months
Contributions summary:Xiao primarily contributed to the development and optimization of the Verl framework, specifically focusing on large language model (LLM) reinforcement learning. Their work includes enhancements to the VLLM rollout process, such as correctly passing parameters for GPU utilization and enabling performance logging. Furthermore, the user implemented sequence parallelism and padding removal optimizations for the SFT trainer, along with adding multi-turn SFT support. The user also incorporated LigerKernel for SFT performance improvements.
swe-train/alan-turing-institute__sktime
Jul 2024 - Jul 2024
Contributions:296 pushes in 1 day
Find and Hire Top DevelopersWeโve 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.