William Lin
Teaching Assistant For CS439 Principles Of Computer Systems at The University of Texas at Austin
Dallas-Fort Worth Metroplex United States
Summary
🤩
Rockstar🎓
Top SchoolWilliam Lin is a systems-focused software engineer and educator with 11 years of experience building and instrumenting backend and MLOps infrastructure while teaching computer systems at UT Austin. He contributes to high-throughput LLM serving (vLLM) work—improving speculative decoding, multi-step scheduling, and backend integrations—bringing practical performance and profiling fixes to production-grade inference stacks. As a long-running CS439 TA and former research assistant in systems and storage, he pairs deep OS and benchmark analysis knowledge with hands-on debugging of student Pintos OS projects. He founded a Rails/Ember startup and has internship experience at VMware, giving him both product and platform perspectives. Based in the Dallas–Fort Worth area, he combines rigorous academic results (3.96 departmental GPA) with open-source engineering that targets latency, memory efficiency, and observability—skills often underappreciated but crucial for scaling ML systems.
10 years of coding experience
2 years of employment as a software developer
Bachelor’s Degree, Computer Science, Departmental GPA: 3.96, Bachelor’s Degree, Computer Science, Departmental GPA: 3.96 at The University of Texas at Austin
English, Chinese
Github Skills (10)
cuda10
pytorch10
mlops10
inference10
python10
llm10
back-end-development9
api9
transformer8
roc7
Programming languages (7)
Github contributions (5)
vllm-project/vllm
May 2024 - Jan 2025
A high-throughput and memory-efficient inference and serving engine for LLMs
Role in this project:
MLOps Engineer & Back-end Developer
MLOps Engineer & Back-end DeveloperContributions:113 reviews, 16 PRs, 181 comments in 8 months
Contributions summary:William primarily focused on extending the vLLM metrics logging API, implementing and improving various functionalities related to speculative decoding, and refactoring worker input preparation for multi-step processing. They also addressed bug fixes related to flashinfer cudagraph capture and torch profiler integration. Furthermore, the user contributed to the addition of the flashinfer backend and multi-step scheduling within the project's core components.
amdcudadeepseekgpthpu
SolitaryThinker/needleman-wunsch
Feb 2019 - Nov 2019
Contributions:69 commits, 43 pushes, 4 branches in 9 months
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.