Hi, I'm Gan Liu
Founding engineer at Citronetic · 3D-vision researcher
At Citronetic we build post-training data and environments for AI agents — simulated workplaces where a model has to read the inbox, dig through chats and docs, write back to the right person, and get graded on the result. Before this I worked on real-time 3D vision, with a paper at CVPR 2026 and the SurgFM-SLAM project. I care about the seams where systems meet.
Research
My stack
About
I started in computer science at SUSTech and spent my early years on real-time 3D vision — depth estimation and scene reconstruction for surgery, where the ground truth is scarce and the scene never holds still. That work became a CVPR 2026 paper and the SurgFM-SLAM project.
These days I'm a founding engineer at Citronetic, where we build post-training data and environments for AI agents: simulated workplaces where a model is handed a real office task — read the inbox, dig through chats and docs, write back to the right person — and its work can be graded and trained on. I've also built the agent infrastructure underneath: unified MCP + REST agent contracts, multi-agent orchestration around a shared state machine, and capability-aware model routing with graceful degradation.
I like owning the whole loop — design it, build it, launch it, iterate. Off the clock you'll find me chasing a good view or hanging out with my cat.
🏜️
🌉
🐱
Experience
Skills
- MCP servers (stdio / http / sse) & tool-calling
- Model routing & graceful degradation
- Multi-agent orchestration & shared-state machines
- Event-driven architecture & event sourcing
- Async job pipelines (submit + poll, queues)
- Guardrails (prompt-injection, DLP, output filtering)
- REST + WebSocket API / SDK design
- Cloudflare Workers / D1 / R2 / KV
- TypeScript / Node.js · Python · C++
- Next.js / React · React Native / Expo
- PyTorch · 3D vision & reconstruction
- Foundation-model adaptation & LoRA fine-tuning
- BigQuery / SQL · PostgreSQL · SQLite (WAL)