Xiaoxi Li (李晓熙)

Xiaoxi Li (李晓熙)

PhD @ RUC

About me

I am currently a 4th year PhD student at the Gaoling School of Artificial Intelligence, Renmin University of China, mentored by Prof. Zhicheng Dou. I earned my bachelor’s degree (2019-2023) at Nankai University.

I’m currently a research intern at ByteDance Seed, focusing on RL scaling for general agents, particularly co-work agent RL and Seed agent RL big runs. I previously interned at Xiaohongshu Inc (the first RedStar program). I have published 30+ papers (8 as first author) in leading journals and conferences, including TPAMI, NeurIPS, ICML, ICLR, ACL, EMNLP, SIGIR, etc. My research and open-source projects have received 3030 citations and 14.0k GitHub stars.

Research Interests：

I aim to build Real-World AI Agents that are reliable, capable, and genuinely helpful in everyday life.

News

  • 🏆 Selected for the CAST Young Talent Development Program for PhD Students (中国科协青年人才培育工程博士生专项).

  • 🚀 Released Agent-Reflex: teaching agents to reflect and recover from tool failures and user corrections. [Code]

  • 🏆 Selected for the CIE-Tencent PhD Research Incentive Program (中国电子学会-腾讯博士生科研激励计划).

  • 🏆 Selected as a BAAI InnoVibe 2026 Most Notable Academic Rising Star (青源InnoVibe-2026最受瞩目学术新星), and gave a talk on “Towards Real-World AI Agents” at the BAAI Conference. [Link]

  • 🚀 Released Agent-World, scaling real-world environment synthesis for evolving general agent intelligence. [Demo]

  • 🎉 DeepAgent has reached 1k stars and selected for Oral Presentation at WWW 2026. A General Reasoning Agent with Scalable Toolsets. [Demo]

  • 🚀 Released OmniGAIA: Towards native omni-modal AI agents. [Demo]

  • 🎉 WebThinker accepted by NeurIPS 2025. (1.2k+ stars, 100+ citations). A powerful deep research agent that can think, search, and write autonomously. [Demo]

  • 🎉 Search-o1 selected for Oral Presentation at EMNLP 2025. (1k+ stars, 200+ citations). The first framework that performs interleaved reasoning and web-search for o1-like reasoning models.

  • 💻 Joined Xiaohongshu as a RedStar research intern, focusing on general agents.

  • 🎉 Four papers accepted by ACL 2025. Looking forward to seeing you in Vienna!

  • 🎉 Our survey on Generative Information Retrieval (GenIR) accepted by ACM TOIS! See more details.

  • 🎉 CorpusLM selected for Oral Presentation at SIGIR 2024. A unified LLM for retrieval and QA.

  • 🎉 Our unified generative framework for retrieval and QA has been accepted by AAAI 2024. See more details.

Educations

  • - Present

    Renmin University of China

    Gaoling School of Artificial Intelligence

    Ph.D. student at the NLPIR Lab, mentored by Prof. Zhicheng Dou.

  • -

    Nankai University

    College of Artificial Intelligence

    B.Eng. in Intelligence Science and Technology (Elite Class).

Research Experiences

  • - Present

    ByteDance Seed, General Agent Team

    Research Intern on RL Scaling for General Agents (Seed-LLM Talent Program)

    Mentors: Wanjun Zhong, Yujia Qin

  • -

    Xiaohongshu, Central Platform Team

    Research Intern on General Agents (RedStar Program)

    Mentors: Wenxiang Jiao, Yuan Lu

Selected Work

LLM/VLM Agents

Reason, Search, and Act
Figure 3: Overview of the DeepAgent framework. The main reasoning model autonomously discovers tools, executes actions, and folds previous memory to restart with structured memories, all within a unified thinking process. The DeepAgent is trained end-to-end wi

DeepAgent: A General Reasoning Agent with Scalable Toolsets

Xiaoxi Li, Wenxiang Jiao, Jiarui Jin, ..., Yuan Lu, Zhicheng Dou Detailed author list

Xiaoxi Li, Wenxiang Jiao, Jiarui Jin, Guanting Dong, Jiajie Jin, Yinuo Wang, Hao Wang, Yutao Zhu, Ji-Rong Wen, Yuan Lu, Zhicheng Dou

WWW 2026 (Oral) CCF A

Most Influential WWW 2026 Papers -- Top 2/950

A general-purpose agent that discovers tools, folds memory, and learns long-horizon interactions with ToolPO.

Paper arXiv Code X (Twitter) REDtechRUC Stars 1.1k Citations 100
Figure 1 Overview of Agent-World (left) and downstream general agent performance (right). The environment- scaling analysis reports the average score across representative subdomains of MCP-Mark, BFCL V4, and 𝜏2-Bench.

Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligence

..., Xiaoxi Li, ..., Ji-Rong Wen, Zhicheng Dou Detailed author list

Guanting Dong, Junting Lu, Junjie Huang, Wanjun Zhong, Longxiang Liu, Shijue Huang, Zhenyu Li, Yang Zhao, Xiaoshuai Song, Xiaoxi Li, Jiajie Jin, Yutao Zhu, Hanbin Wang, Fangyu Lei, Qinyu Luo, Mingyang Chen, Zehui Chen, Jiazhan Feng, Ji-Rong Wen, Zhicheng Dou

Preprint, Under Review

Project Page Paper arXiv Code 机器之心 Citations 40

Retrieval-Augmented LLMs

Retrieve and Generate

Reward Modeling and Machine Learning

Learn and Align

View this figure in the paper ↗

Selected Awards

  • 2026: CIE-Tencent Hunyuan PhD Research Incentive Program (中国电子学会-腾讯混元博士生科研激励计划，10w/人).
  • 2026: BAAI InnoVibe Most Notable Academic Rising Star (青源InnoVibe最受瞩目学术新星，全球25人).
  • 2026: Young Talent Development Program for PhD Students, CAST (中国科协青年人才培育工程博士生专项).
  • 2026: National Scholarship for PhD Students, Renmin University of China (中国人民大学博士生国家奖学金; Top 1%).

Invited Talks

  • 2026.06: “Towards Real-World AI Agents: Technical Evolution and Representative Work”, BAAI Conference. [Link]
  • 2026.05: “Reason, Search, and Act: Towards Real-World AI Agents”, MLNLP Community. [Link]
  • 2026.03: “Reason, Search, and Act: Towards Real-World AI Agents”, CCIR PhD Forum.
  • 2025.04: “WebThinker: Empowering Large Reasoning Models with Deep Research Capability”, MLNLP Community. [Link]

Academic Services

  • PC Member: ACL Rolling Review, NeurIPS, ICML, ICLR, SIGIR, AAAI (Silver Reviewer Award @ ICML’26)
  • Journal Reviewer: ACM Computing Surveys, TMLR, TOIS