About

I am a master's student at the School of Artificial Intelligence and Data Science, University of Science and Technology of China, in the LDS Lab. From September 2025 to October 2026 I was a research intern at Upwork. I work on AI agents and large language models: long-term memory for agents, whether coding agents can clarify what clients actually want, and how research agents can keep their direction over long tasks.

Before agents, I worked on generative recommendation. I also build open-source tools for research, including Open-Research-Agent.

NowSWE-Clar and MESA (both co-first author) are under review at WWW 2027, and UniGRec at ICLR 2027. Looking for internships on agents and LLMs.

  • AI agents
  • Large language models
  • Generative recommendation
  • Agent-driven quant research (self-study)

News

Publications

MESA running example: a user's matters over many sessions, a memory of pieces ranked by similarity, and MESA where each matter has an owner agent

MESA: Long-Term Agent Memory as an Emergent Society of Agents

WWW 2027Under reviewCo-first author

A long-term memory in which each matter the user pursues is owned by an agent that keeps its situation and decides whether a request concerns it; the society of agents grows from the user's history.

Experience

  • Upwork

    Research InternNasdaq-listed global freelance marketplace

    • SWE-ClarWWW 2027 Under reviewCo-first author

      A benchmark built from real client jobs that tests whether coding agents clarify requirements and deliver working software.

    • UniGRecICLR 2027 Under reviewFirst author

      Differentiable soft identifiers that optimise the item tokenizer and the generative recommender end to end.

    • SODARecSys 2026 Short Paper

      Distribution-level contrastive supervision that gives generative recommendation finer soft semantics.

    2025.09 – 2026.10

Projects

Open source · in development

Open-Research-Agent

An open-source research agent, in development: a persistent research map keeps its direction and an independent advisor reviews each round. I am now studying which parts of its own history a long-horizon agent should see at each step.

In progress

Contract Review Agent

Exploring how agents can review a contract on their own, finding and fixing the errors in it. A team project for a legal-AI competition.

Products

Live on the web

Dougao

A fuse-bead pattern tool: turns a photo or an AI prompt into an editable pattern, with colour merging, bead inventory and offline-first cloud sync.

Live · open source

ORA Paper Reader

An agent-assisted paper reader: import a PDF or arXiv paper and get a guide whose every point links back to the page. The agent also helps take notes and recommends papers that match your research interests.

Education

  • University of Science and Technology of China

    M.Sc. in Big Data Technology and Engineering, School of AI and Data Science, LDS Lab

    2026 – Present
  • Lanzhou University

    B.Eng. in Data Science and Big Data Technology

    GPA 4.18, average 91.77, ranked 1st of 37 in the major; CET-6 506

    National Scholarship · Merit Student of Gansu Province · MCM Honorable Mention · China Collegiate Computing Design Contest, provincial first prize · APMCM third prize

    2022.09 – 2026.06