Jiawei Liu
I am a final-year Ph.D. student at University of Illinois Urbana-Champaign, working with Lingming Zhang. My research explores how Software Engineering and Programming Language can benefit and benefit from Machine Learning and its systems, with a focus on improving software reliability and developer productivity.
Software Engineering with Language Models
- Code evaluation: correctness [EvalPlus], efficiency [EvalPerf], without hard verifiers [CodeFavor]
- Training models [StarCoder2] to reason [Code-R1] and follow diverse instructions [Magicoder]
- Code editing should be real-time and can be largely accelerated by multi-layer speculation [Blazedit]
Software Engineering for ML Systems
My research has been generously supported by Amazon PhD Fellowship, Illinois Innovation Award, Yee Memorial Fellowship, and grants from Amazon and OpenAI. My work (i) finds 300+ critical bugs automatically in ML systems like PyTorch, winning ACM SIGSOFT Distinguished Paper Award and Distinguished Artifact Award, and (ii) builds LLMs and evaluators for code with 1M+ downloads and wide industrial adoptions.
📰 Some recent coding: R1 for Code Generation and Speculative Code Editing.
ResearchShow More
- Proc. ACM Softw. Eng. 2 (ISSTA). Jun 2025
- Forty-first International Conference on Machine Learning. Jun 2024Adopted by Meta Llama 3.1, Google CodeGemma, and IBM Granite
- arXiv preprint arXiv:2402.19173. Jun 2024
- Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering. Jun 2023🏆 ACM SIGSOFT Distinguished Paper Award
- Proceedings of the ACM on Programming Languages 6 (OOPSLA1). Apr 2022
Awards & Honors
Illinois Innovation Award ($20K) 2025
Amazon AICE Ph.D. Fellowship ($70K) 2025
Jane Street Fellowship Honorable Mention 2025
Proposal, Amazon Nova AI Challenge ($250K) 2024
OpenAI Researcher Access Program 2024
Machine Learning and Systems Rising Stars 2024
Warren W. Yee Memorial Fellowship 2024
Service
Organizing: LLM4Code@ICSE (Publicity Chair)
Invited Talk
NLP+SE Seminar, UT Austin: Smelling the Quality of LLM-generated Code Mar 2025
Programming Systems, Uber: Evaluating LLMs for Correct & Efficient Code Generation Sept 2024
ARiSE Lab, Columbia University: Simplify the Making of Great Software in the ML Era April 2024
Snowflake GenAI: Rigorous Evaluation of LLMs for Code (Slides) Feb 2024
AST Lab, ETH Zürich: Generating Test-Cases for ML Compilers (Slides) Jan 2024
GAI4SE, NC State University: LLMs for Software Testing (Guest Lecture) Nov 2023
Apache TVM Conference: Automating DL Compiler Bug Finding with NNSmith Mar 2023
SAMPL, University of Washington: Coverage-Guided Tensor Compiler Fuzzing May 2022