About
I'm an open-source developer and researcher focused on Agent RL, LLM post-training, and AI agents. Projects I build and maintain have earned 100,000+ GitHub stars in total — including Happy-LLM, self-llm, and agentic-rl-lab.
Currently I'm a researcher at Emotion Machine Lab, where I also work across developer growth and the open-source community. Previously I was a research assistant at Westlake University, AGI Lab — where my first-author paper Hard to Read, Easy to Jailbreak was published at Findings of ACL 2026 — and at Yunqi Academy of Engineering.
- Agent RL
- LLM Post-training
- AI Agents
- Multimodal Safety
Experience
-
Present
Researcher & Growth Operations · Emotion Machine Lab
- Research on Agent RL, LLM post-training, and agentic systems.
- Developer growth, community operations, and the open-source ecosystem.
-
Research Assistant
Westlake University · AGI Lab
- Multimodal model behavior and safety, with Prof. Chi Zhang.
- First-authored Hard to Read, Easy to Jailbreak, Findings of ACL 2026.
-
2024.06 — 2024.08
Research Assistant · Yunqi Academy of Engineering
- LLM applications in urban governance.
- Co-authored What can LLM tell us about cities?
Selected Publications
- Hard to Read, Easy to Jailbreak: How Visual Degradation Bypasses MLLM Safety Alignment Findings of ACL 2026 · First author
- What can LLM tell us about cities? arXiv · 2024
- A Raster Voronoi Diagram Generating Algorithm Using Edge Attribution and Bilateral Scanning Geomatics and Information Science of Wuhan University · 2024
Projects
-
>_
Happy-LLM
A from-scratch guide to LLM fundamentals and implementation.
-
>_
self-llm
A practical guide to deploying and fine-tuning open-source LLMs.
-
>_
tiny-universe
Hands-on implementations of RAG, agents, evaluation, and other LLM systems.
-
>_
agentic-rl-lab
Reproductions and studies of RL algorithms for LLM agents.
-
>_
paper-insight
An AI-powered platform for paper analysis and research workflows.
-
>_
huanhuan-chat
A ChatGLM-based character chatbot fine-tuned on dialogue from Empresses in the Palace.
Notes
A wall of sticky notes — what I'm up to and random thoughts. On desktop you can drag them around; the arrangement is saved in your browser.