Faculty Profile: Prof. Lingming Zhang — UIUC ISE Lab
Faculty Profile: Prof. Lingming Zhang — UIUC ISE Lab
Position: Associate Professor Institution: UIUC, Siebel School of Computing and Data Science Lab: ISE Lab (Intelligent Software Engineering) — ise-uiuc.github.io Website: https://lingming.cs.illinois.edu/ Report date: 2026-06-12
Research Focus
Software Engineering + LLM intersection: code generation (SFT/OSS-Instruct), automated software testing/repair, software engineering agents (SWE-bench), RL training for code agents (SWE-RL, self-play RL).
Academic Profile
- ACM SIGSOFT Early Career Researcher Award
- NSF CAREER Award
- UIUC Dean’s Award for Excellence in Research
- 2025 Google Academic Research Award (LLM agent vulnerability detection)
- Industry research awards: Alibaba, Google, Meta, Samsung, Kwai Inc.
- Program co-chair: ASE 2025, LLM4Code 2025
Key Publications
| Paper | Venue | Relevance | |——-|——-|———–| | Magicoder / OSS-Instruct | ICML’24 | SFT data pipeline for code LLMs; adopted by Meta Llama 3, Qwen2.5-Coder, CodeGemma, IBM Granite | | SelfCodeAlign | NeurIPS’24 | Self-alignment for code LLMs | | Agentless | 2024 | SE agent framework demystifying LLM-based SE agents | | SWE-RL | NeurIPS’25 | RL training for software engineering agents (w/ Meta FAIR) | | Self-Play SWE-RL (SSR) | ICML’26 | Self-play RL → training toward superintelligent software agents |
Known Student Outcomes
| Student | Status | Placement / Signal | |———|——–|——————-| | Yuxiang Wei | PhD candidate (graduating ~2026) | Meta FAIR, CodeGen/CodeLlama team (student researcher 2024–2025); lead on SWE-RL, SSR, Magicoder | | Yinlin Deng | PhD student | Agentless co-lead | | Chunqiu Steven Xia | PhD student | Agentless first author | | Zhe Wang | PhD student | Active | | Yifeng Ding | PhD student (from Fall 2022) | Active |
Frontier placement signal: Yuxiang Wei → Meta FAIR (CodeGen team). One confirmed frontier placement in progress.
Fit with Weijia Zhang
| Dimension | Assessment |
|---|---|
| SFT methodology / data pipeline | ✅ Direct (Magicoder OSS-Instruct mirrors Weijia’s TextAnalysisSFT at MSRA) |
| RL post-training for agents | ✅ Direct (SWE-RL, Self-Play SWE-RL) |
| Agentic AI framework design | ✅ Agentless, SWE-bench ecosystem |
| Code LLMs specifically | ✅ Core focus |
| GUI / VLM agents | ❌ No visual modality |
| General NLP / dialogue | ❌ Domain locked to code and SE |
| Multimodal / embodied | ❌ Not his area |
Risks
- Domain specificity: All work is code/SE — Weijia’s interests span broader agentic AI (GUI, multimodal, dialogue)
- Pivot required: Joining would mean committing to code agent direction
- Lab size and bandwidth: Medium-sized lab, PI is productive and hands-on based on publication rate
Verdict
Method match is excellent (SFT data pipelines + RL training + agent frameworks), but the application domain is firmly code/software engineering. Ideal if Weijia decides to specialize in code agents / SE agents. If staying with GUI agent / general agentic AI direction, he is a strong collaborative peer rather than primary advisor.