Xuan (Allen) Zhao (@zx0125) profile picture

MSc at ETH, working on medical AI

Joined June 2021
Photo by Xuan (Allen) Zhao (@zx0125)Photo by Xuan (Allen) Zhao (@zx0125)Photo by Xuan (Allen) Zhao (@zx0125)Photo by Xuan (Allen) Zhao (@zx0125)Photo by Xuan (Allen) Zhao (@zx0125)Photo by Xuan (Allen) Zhao (@zx0125)
Xuan (Allen) Zhao retweeted
2.5 years ago we released AgentClinic, an open benchmark where doctor agents diagnosed patient agents through chat and tools. Now we ask, can we take this to the next level and give those agents a 3D embodied hospital? What started as a fun weekend hack became GTAgentClinic, an open-source 3D hospital full of agents - for agents and humans alike. Sound on 🔊
11
9
4
71
117,627
II’ll be presenting AgentGUI at EMNLP 2026 in Budapest! 📌 Hall H, Demo Session A ⏳ Sunday (Oct 25), 11:00-12:30pm AgentGUI is a pixel-art interface built for humans to make observing and steering agents efficient, insightful, and fun. Let’s chat! (1/5)
1
1
1
9
862
With the popularization of AI agents, human insights into agent traces is an important step towards trust and improvement. Come take a look at how we do this, and looking forward to chatting! (4/5)
1
23
Preprint: lnkd.in/ef-4uU-g Github: lnkd.in/ejF6cGCT Project Website: lnkd.in/emK-8K7P Demo video: lnkd.in/e_CHEDJE Big thanks to the ETH Medical AI Lab, and @Michael_D_Moor for the supervision and support!
37
AgentGUI has been accepted to the #EMNLP 2026 System Demonstrations track! It started as a one-day lab hackathon, got a new release during the @Anthropic x @ETH AI Sprint, and is now a conference paper. (1/7)
1
2
6
304
Also thanks a lot to the folks from ETH Medical AI Lab Ashraya Indrakanti, Jiang Yulun, Johann Lieberwirth, and Lorenzo Andrea Menarbin for feedback and bug hunting along the way. (6/7)
1
1
57
Running agents overnight, we wake up to results we want to believe in, but find it hard to so without going through the long and winding road of agent transcripts. For tasks that are not easily verifiable, human insights into agent actions is essential for trust and progress. 1/n
1
2
1
7
1,013
Users understand agent trajectories 38% faster, without sacrificing accuracy:
4
80
A new release (v1.1.0) of AgentGUI is available after the Anthropic-ETH AI Sprint! AgentGUI is a user-friendly interface for observing and steering long-running agents. This release extends support beyond Hermes agents, and improves support for complex research tasks. 1/3
1
1
2
1,004
Release highlights: * Use and visualize a Claude agent with your Anthropic subscription, not requiring an API billing key * Auto-spawned subagent visualizations * Let your agent train AI models efficiently by running on remote GPU servers 2/3
1
1
64