AdsMind · CatDT · Laboratory safety
Research milestones · Character cabinet · Developer vignette
📑 Table of Contents (Click to expand)
I am an undergraduate at The Hong Kong University of Science and Technology (HKUST), majoring in Computer Science with a minor in Chemistry. I was an exchange student at EPFL, Switzerland and took virtual coursework at SJTU. My research lies at the intersection of AI and the physical sciences (AI4Science) — building multi-agent systems that autonomously discover and validate catalytic materials.
As the Co-Founder & Vice President of the HKUST AI for Chemistry Club, I am deeply passionate about bridging the gap between computer science algorithms and wet-lab chemical intuition, fostering a community of interdisciplinary innovators.
- 🔬 Research: Multi-Agent Systems for autonomous catalyst discovery, LLM-driven scientific workflows, Embodied AI for lab safety.
🏅 Honors, service & languages
- 🏆 Honors: HKUST SENG Dean's List ×5 · HKSAR Government Scholarship Fund — Scholarship for Outstanding Performance ×3 · Hong Kong, China-APEC Scholarship ×3.
- 🏅 Competition Awards: Finalist Award (Team Category), HKUST GenAI Hackathon Competition on a Putonghua Web Tool · First Prize (Theory Track), 5th National University Students' Data Analysis Popular Science Knowledge Competition Series.
- 📝 Service: Reviewer for MLST (IOP), NeurIPS'26 AI4S Workshop, ICML'26 AI4S Workshop.
- 🌏 Languages: Mandarin (native) · English (professional) · Japanese (approximately JLPT N3 level) · French (CEFR A1) · Cantonese (basic).
- ⚡ Off-duty: 5 human languages, 7 programming languages, and occasionally building games in Godot & Unity.
- 💬 Let’s connect: AI for Science, AI for Chemistry, AutoResearch, and open-source collaboration — email me.
🗓️ Recent research updates (click to expand)
2026-01-28 · Merged Planner–Critic PR · 2026-05-04 · Community report · 2026-06-03 · CatDT v1 · 2026-06-17 · AdsMind v1 · 2026-07-09 · CatDT v2
AdsMind 🚀
🎞️ Animated research loop · workflow schematic · click to toggle

A conceptual loop illustration; it does not depict a recorded run or measured timing.
Finding stable adsorption configurations on heterogeneous catalyst surfaces has long been a laborious trial-and-error process. AdsMind automates this with a multi-agent closed loop: LLM agents propose candidate structures, a machine-learning force field relaxes them, and a physics-grounded verifier self-corrects errors — autonomous correction within the configured workflow.
Reported preprint results and benchmark scope.
My contribution: Led the closed-loop architecture development, built the benchmark framework, and conducted cross-backend evaluation. Code history.
Try it: Start with the supplied catalyst slabs and an adsorbate SMILES; inspect the run report, structure files, and rendered configuration (when available).
CatDT 🧪
A self-evolving multi-agent digital twin for autonomous heterogeneous catalyst discovery, coordinating surface reconstruction, reaction-pathway analysis, and microkinetics from a bulk crystal and a natural-language reaction description.
Reported preprint results and benchmark scope.
My contribution: Co-author and research contributor to CatDT — paper authorship · system implementation.
Merged upstream PR · NetSentinel contribution record.
| # | Paper | Status |
|---|---|---|
| 1 | AdsMind: A Physics-Grounded Multi-Agent System for Self-Correcting Discovery of Adsorption Configurations on Heterogeneous Catalyst Surfaces | Under review · arXiv:2606.19152 |
| 2 | Autonomous Heterogeneous Catalyst Discovery with a Self-Evolving Multi-Agent Digital Twin | Under review · arXiv:2606.05050 |
| 3 | From Knowledge to Action: Outcomes of the 2025 Large Language Model (LLM) Hackathon for Applications in Materials Science and Chemistry | arXiv:2605.03205 |
| Project | Paper | Code | Data / examples | My work |
|---|---|---|---|---|
| AdsMind | arXiv | GitHub | Public directory | Project overview |
| CatDT | arXiv | GitHub | Public directory | Project overview |
| From Knowledge to Action | arXiv | — | — | Paper authors |
🔬 Public repository examples · click to open
AdsMind: a projection of 54 committed Pt coordinates, showing a public input structure rather than a new optimization result. Source · XYZ
CatDT: committed AdsorbDiff module example images, showing the initial and diffused structures. These are not a new run or an end-to-end validation of CatDT. Source
💬 Ask a public research question · ✉️ Private collaboration
🖼️ Research visual gallery · papers & code
AdsMind PDF · Code · CatDT PDF · Code
Visual summaries of existing preprints; these are not conference posters or published book covers. Public talks, models and posters can be added when available.
Final Year Project — HKUST (2026 – 2027 expected)
Embodied Multi-Agent System for High-Stakes Environment Safety Advisors: Prof. Chaojian Li & Prof. Lixue Cheng
The project aims to transfer autonomous driving architectures to edge-deployed labs, combining multi-agent reasoning, multimodal sensor fusion, and hardware-in-the-loop (HIL) control for laboratory safety.
UGRA — AI4PhysSci Lab, HKUST (02/2026 – present)
Supervisor: Prof. Lixue Cheng
- Led the development of AdsMind: closed-loop architecture integrating generative agent proposals with machine-learning force-field (MLFF) relaxation feedback and self-correction, with DFT validation on representative systems.
- Built a systematic multi-agent benchmark for adsorption structure prediction across diverse catalyst surfaces.
- Contributed to CatDT: self-evolving multi-agent digital-twin framework for autonomous heterogeneous catalyst discovery.
Project Student — LIAC, ISIC, EPFL (09/2025 – 01/2026)
Supervisors: Prof. Philippe Schwaller & Dr. Edvin Fako
Developed agentic simulation methods combining language-model reasoning with atomistic simulation backends. This semester project established the EPFL–HKUST collaboration that produced AdsMind.
UROP — HKUST, Prof. Xiaojuan Ma (02 – 05/2025)
User Experience & Visual Representation in Virtual Reality
UROP — HKUST, Prof. Raymond Chi-Wing Wong (06 – 12/2024)
Knowledge Discovery Over Database
| Institution | Program | Period |
|---|---|---|
| HKUST | BEng Computer Science · Minor in Chemistry · CGA 3.999/4.3 | 09/2023 – present |
| EPFL, Switzerland | Exchange, School of Computer and Communication Science | 09/2025 – 02/2026 |
| SJTU (APRU Virtual) | MSE2602 Materials Chemistry | 02 – 06/2025 |
| Role | Organization | Period |
|---|---|---|
| Computer Vision Algorithm Intern | Sansen Well (Shanghai) Robotics Co., Ltd. | 06 – 08/2025 |
| UGTA: COMP2011, CSE Programming Commons | HKUST | 2024 – 2026 |
| Co-Founder & Vice President | HKUST AI for Chemistry Club | 09/2026 – present |
| Student Representative, SENG Mainland UG Recruitment | HKUST, Guangzhou | 06 – 07/2026 |
In use: Python · PyTorch · NumPy · LangChain / LangGraph · RDKit / ASE.
Exploring: Fine-tuning · Embodied AI · Laboratory safety.
| Research workflows | Development & other experience |
|---|---|
|
AI / ML / LLM Scientific Computing |
🤝 Browse bookmarks, activity and repository contributors
Bookmarks are grouped by repository language. Contributor attribution comes from public commits and does not imply paper authorship or endorsement. Activity covers the latest 100 public events.
📖 Open the latest notes · direct file links
| Course & theory notes | Hands-on tool notes |
|---|---|
| GenAI · Hung-yi Lee | RDKit |
| NLP | Fine-tuning |
| Quantum Information | Chinese-language personal study notes; see each repository for setup. |
- 元素周期律 + 元素周期表 · 2021-09-02
- 酸碱理论 · 2021-09-01
- 氮族元素——磷 · 2021-08-28
Selected Chinese study notes from 2021; the decorative covers are original artwork, not article thumbnails.
Ooyodo · Agent-and-Wolf · Ibuki pet
🌸 Character cabinet · click to open
Nagato uses the approved icon; Ibuki uses a frame from the project spritesheet; Holo is represented by a wolf-and-wheat emblem.
| Character | Original work | Connection here |
|---|---|---|
| Nagato | Azur Lane | GitHub avatar and profile visual identity |
| Ibuki | Azur Lane | Animated desktop companion |
| Holo | Spice and Wolf | ReAct role-playing agent |
✏️ Add a topic · 🔀 Shuffle · 🤖 Update status
Sign in to submit a short scientific topic. Your newest valid submission replaces earlier ones; close it to withdraw, or reopen it to vote again. Updates run after submissions; GitHub image caching may delay display.
🖼️ Quarterly cloud gallery · click to open
Each quarter retains its latest real-vote snapshot; example words are excluded. Closing all current votes clears this quarter’s snapshot. Older snapshots show historical counts.
Repositories · Followers · Update status
⌛ Weekly coding activity · public data
Only public WakaTime records are used. Until connected, no hours are shown; the language chart above describes repository contents.








