PhD Student at Tsinghua University.
My research lies at the intersection of LLM/VLM Evaluation & Alignment, Multi-Agent Systems, and Urban AI. I design benchmarks and training-free alignment methods for large generative models, and build global-scale urban datasets. Passionate about building AI productivity tools.
For AI Agents — give AI memory and workspace:
| Project | What It Does |
|---|---|
| MemoMind | Memory system for AI agents — persistent, evolving knowledge graph across sessions |
| Zync | IDE for AI agents — orchestrate multiple agents (Claude Code, Codex, Aider) in parallel |
For Humans — make AI work for you:
| Project | What It Does |
|---|---|
| Recall | AI conversation manager — browse, search, and never lose a conversation |
| HumanSkillTree | AI-augmented skill system — use AI to enhance human learning and capabilities |
| Usage Monitor | Real-time Claude Code quota dashboard — 5h/7d usage, reset countdowns, SSH status |
| Desktop Controller | AI Computer Use for Claude Code — control WeChat, DingTalk, any app via Win32 API + Playwright |
Connected Everything — one voice, all devices:
| Project | What It Does |
|---|---|
| DeviceMesh | AI + Tailscale + SSH — control all your devices with natural language from anywhere |
| DigitalMe | Chat App → AI Agent → Terminal. Control AI agents via Feishu/Telegram/Slack |
DeviceMesh for infrastructure. DigitalMe for messaging. Together, one voice controls everything.
Evolved from AnyTerm (browser-based remote terminal) — replaced by AI-native device control.
Key works:
🌐 Academic Homepage · 🎓 Google Scholar · 📧 Email · 💬 QQ: 575860760