01

About

关于

I am a PhD student at Westlake University and a Top Intern Program intern at Awomo, working on agentic systems, reliable reasoning, and scientific knowledge.

我目前是西湖大学博士研究生,也是 Awomo Top Intern Program 实习生,主要研究智能体系统、可靠推理与科学知识。

I build systems that turn evidence and environments into action: THE-Tree reconstructs scientific evolution; SimForge, Unravel, and Physical RSI connect 3D worlds, rollouts, feedback, and improvement. AI for Science is an important application domain, alongside molecular, medical, and embodied settings.

我构建把证据与环境转化为行动的系统:THE-Tree 重建科学演化,SimForge、Unravel 与 Physical RSI 连接三维世界、智能体 rollout、反馈与改进。AI for Science 是重要应用场景,同时覆盖分子、医学与具身智能。

  • Agentic Systems
  • 智能体系统
  • Reliable Reasoning
  • 可靠推理
  • Scientific Knowledge
  • 科学知识
  • Physical AI
  • 具身智能
  • 3D Simulation
  • 三维仿真
  • Multimodal Evaluation
  • 多模态评测
  • AI for Science
  • 科学智能
  • Cross-disciplinary Practice
  • 跨界实践
  • AI for Education
  • AI 教育
  • Cultural Inquiry
  • 文化探究
  • Community Organizing
  • 社群组织
  • Public-interest Technology
  • 公益科技
NOW目前Awomo · Westlake UniversityAwomo · 西湖大学Top Intern Program · PhD studentTop Intern Program · 博士研究生
COMMUNITY社区Tencent Dayan · Datawhale · OpenI · PPDE腾讯大衍 · Datawhale · OpenI · PPDE
02

Community & recognition

社区与认可

Research is only one part of my work. I also learn, teach, and contribute through the developer communities around open models and scientific AI.

研究只是我工作的一部分。我也在开源模型与科学智能社区中学习、分享并持续贡献。

03

Beyond research

科研之外

I care about how technical knowledge travels across communities. My education and cultural work connects people, institutions, and local contexts.

我也关注知识如何进入真实的教育与文化场景,通过连接个人、机构与地方社区,探索教育、文化与公益科技的可能性。

TEACHING & EDUCATION支教与教育

Jingmai Mountain · Education outreach

景迈山 · 教育实践

During a visit to Jingmai Mountain in Yunnan, I noticed local education needs and initiated contact with Zhejiang University's Xinghuochao volunteer teaching team. I coordinated with local government, entrepreneurs, and villagers to arrange teaching spaces, accommodation, and local support, while exploring how AI for Education could serve remote communities.

在云南景迈山旅行期间关注到当地教育需求后,我主动联系浙大星火潮支教团队,并协调当地政府、企业家与村民,落实授课场地、住宿和在地支持,同时探索 AI for Education 服务边远地区的可能性。我也尝试联系西湖大学科教团,寻找更长期的教育合作路径。

CULTURAL COMMUNICATION文化交流

Dunhuang · Humanities on campus

敦煌 · 校园人文实践

Through the Future Humanity Institute and personal outreach, I invited a Dunhuang researcher and teacher, Shi Laoshi, to engage with Westlake University and explore Dunhuang culture, humanities, and nature.

我通过未来人类研究院及个人联系,邀请敦煌石老师走进西湖大学,围绕敦煌文化、人文与自然展开交流和深入探究。

CULTURAL INQUIRY文化探究

Confucianism · Buddhism · Taoism

儒 · 释 · 道

I am deeply interested in Chinese traditional culture, especially the dialogue among Confucianism, Buddhism, and Taoism. This curiosity shapes how I think about education, technology, and the relationship between people and place.

我长期关注中国传统文化,尤其是儒、释、道三家的思想脉络与彼此对话。这种文化探究也影响我对教育、技术以及人与地方关系的理解。

04

Research focus

研究方向

A

Agentic systems & Physical RSI

智能体系统与 Physical RSI

Planning, tool use, simulation, rollout feedback, and recursive self-improvement for agents that learn to act.

围绕规划、工具调用、仿真、rollout 反馈与递归式自我改进,研究能够学习行动的智能体。

↗
B

Reliable reasoning & evidence

可靠推理与证据验证

Grounding model outputs in evidence, tracing claims, and diagnosing where reasoning fails.

让模型输出建立在证据之上,追踪论断来源,并诊断推理失败的位置。

↗
C

Scientific knowledge & evolution

科学知识与技术演化

THE-Tree turns literature into time-aware, citation-linked paths for verification and forecasting.

THE-Tree 将论文组织成带时间与引用证据的演化路径,用于验证与趋势预测。

↗
D

Multimodal evaluation in science

科学场景中的多模态评测

Medical-image quality, molecular grounding, and 3D scene understanding beyond single scalar scores.

关注医学影像质量、分子 grounding 与三维场景理解,超越单一分数评测。

↗
05

Experience

经历

Mar 2026 — Present2026 年 3 月 — 至今

Awomo

Top Intern Program · Lead, 3D simulation data construction

Top Intern Program · 3D 仿真数据构建负责人

Lead SimForge and Haness pipelines for simulation-ready scenes, robot-learning demonstrations, and the Physical RSI loop connecting environments, rollouts, feedback, and improvement.

负责 SimForge 与 Haness,构建可用于仿真的三维场景、机器人学习示范数据,并搭建连接环境、rollout、反馈与改进的 Physical RSI 闭环。

2025 — Present2025 — 至今

Autolab · Westlake University西湖大学 · Autolab

PhD student · agentic systems / AI for Science

博士研究生 · 智能体系统 / AI for Science

Research on agentic systems, reliable reasoning, multimodal understanding, and scientific applications.

研究智能体系统、可靠推理、多模态理解与科学应用。

Nov 2024 — Feb 20262024 年 11 月 — 2026 年 2 月

Shanghai AI Laboratory上海人工智能实验室

Researcher · scientific knowledge and evaluation

研究员 · 科学知识与评测

Led THE-Tree and contributed to MedQ-Bench, a multimodal benchmark for medical-image quality and reasoning.

负责 THE-Tree,并参与面向医学影像质量与推理的多模态评测基准 MedQ-Bench。

Dec 2023 — Jun 20242023 年 12 月 — 2024 年 6 月

Roche

AI-based drug design intern

AI 药物设计实习生

Compound-fingerprint modeling and LangChain + GPT-4 workflows for pharmaceutical patent analysis.

开发化合物指纹模型,并用 LangChain + GPT-4 提升药物专利分析效率。

06

Selected work

代表工作

01arXiv · 2026arXiv · 2026

SimForge · Awomo-SimDataEngine ↗

An agentic simulation-ready world pipeline connecting ISArt assets, Unravel reconstruction, SimForge text-to-scene generation, validation, and PolicyForge demonstrations.

连接 ISArt 资产、Unravel 重建、SimForge 文本生成场景、验证与 PolicyForge 示范数据的智能体仿真世界系统。

02ICLR 2027 submissionICLR 2027 投稿

RSI-CUA

A Physical RSI framework for computer-use agents: environments, rollouts, feedback, and iterative improvement in an endless training ground.

面向计算机使用智能体的 Physical RSI 框架:在持续训练环境中连接环境、rollout、反馈与迭代改进。

03Findings of EMNLP 2026EMNLP 2026 Findings

THE-Tree ↗

A temporal scientific knowledge-graph system for reconstructing research evolution, grounding verification in evidence, and supporting trend forecasting.

面向科学研究演化重建的时序知识图谱系统,用证据支撑科学验证,并辅助趋势预测。

04EMNLP 2026 Oral · Outstanding Paper AwardEMNLP 2026 Oral · Outstanding Paper Award

MedQ-Bench ↗

A perception–reasoning benchmark for medical-image quality across 5 modalities and 40+ attributes, with 2,600 perception queries and 708 reasoning assessments.

面向医学影像质量的感知—推理评测基准,覆盖 5 种模态、40 多项属性、2,600 个感知问题与 708 个推理评估。

05arXiv · 2024

PatentAgent ↗

A multi-module LLM agent for patent QA, image-to-molecule extraction, and central-compound identification.

面向专利问答、结构图像转分子与核心化合物识别的多模块大模型智能体。

06arXiv · 2026

R-GroundBench ↗

A diagnostic benchmark for reliable R-group grounding in Markush molecular editing.

面向 Markush 分子编辑中可靠 R 基团 grounding 的诊断型评测基准。

07

Internship targets

求职方向

LOOKING FOR正在寻找

Research internships in agentic systems, physical AI, reliable reasoning, and scientific evaluation.

寻找智能体系统、具身智能、可靠推理与科学评测方向的研究型实习。

Agentic Systems智能体系统Physical RSIPhysical RSI3D Simulation三维仿真Multimodal Evaluation多模态评测AI for Science科学智能LLM post-training大模型后训练
Open the official-link watchlist查看官方入口清单 ↗