Hi there!👋 I am LIU Mugeng, a Ph.D. candidate at the School of Computer Science, Peking University, advised by Prof. MEI Hong, Prof. LIU Xuanzhe, and Prof. MA Yun. Previously, I received my B.S. in Computer Science from Peking University.

My research lies at the intersection of artificial intelligence and software engineering/systems. My current core agenda is agent harness: the systems layer around LLM agents that turns model capability into reliable real-world applications.

I have published 14+ papers at leading venues, spanning ICML, ICLR, EMNLP, EACL, SIGIR, WWW, ICSE, FSE, and TOSEM. Beyond publications, I lead and contribute to real-world agent and AI systems, mainly in ThinkSync AI product systems such as OpenWebClaw, SparkBox, WebAct, and ThinkX, and deployed educational AI systems such as ATMOS.

🔍 Research Interests

🔥 News

• May. 2026: 🎉🎉🎉 Our paper on flexible agentic web service interface was accepted by ICML 2026!
• Mar. 2026: 🎉🎉🎉 Our ThinkSync-AI team open-sourced OpenWebClaw, a web-native AI agent built for personalized and long-running web tasks, with self-revolution driven by continuously evolving skills, memories, and experience.
• Mar. 2026: 🎉🎉🎉 Our paper “From Task to Tutorial” was accepted by FSE 2026!
• Jan. 2026: 🎉🎉🎉 Our paper “ROGA” on efficient self-envolving office agent was accepted by ICLR 2026!
• Jan. 2026: 🎉🎉🎉 Our work “WAB” on efficient agentic web runtime was accepted by FAAW @ WWW 2026!
• Jan. 2026: 🎉🎉🎉 Our paper “LaTune” was accepted by WWW 2026!
• Jan. 2026: 🎉🎉🎉 Our paper “DRAGON” was accepted by EACL 2026!
• Jan. 2026: 🎉🎉🎉 Our paper on bugs in LLM inference engines was accepted by TOSEM 2026!
• Dec. 2025: 🎉🎉🎉 Our empirical study on WebGPU was accepted by ICSE 2026!
• Aug. 2025: 🎉🎉🎉 Our paper “SheetDesigner” was accepted by EMNLP 2025! This work was done during my internship at Microsoft!
• Apr. 2025: 🎉🎉🎉 Our paper “WebANNS” on efficient in-browser RAG was accepted by SIGIR 2025!
• Jan. 2025: 🎉🎉🎉 Our paper “WeInfer” was accepted by WWW 2025!
• Jan. 2025: 🎉🎉🎉 Our survey paper on WebAssembly runtimes was accepted by TOSEM 2025!
• Dec. 2024: 🎉🎉🎉 Our paper on Wasm-based container runtime was accepted by TOSEM 2025!
• Mar. 2024: 🎉🎉🎉 Our paper on research artifacts in SE publications was accepted by JSS 2024!

📖 Educations

• [Sep. 2022 - Jul. 2027] Ph.D., School of Computer Science, Peking University
• [Sep. 2018 - Jul. 2022] B.S., School of Electronics Engineering and Computer Science, Peking University

📝 Selected Publications

• Liu, Mugeng, et al. “Beyond Static Endpoints: Tool Programs as an Interface for Flexible Agentic Web Services.” (ICML 2026, CCF-A).
• Liu, Mugeng, et al. “ROGA: Scaling Generalist Agents for Office Productivity Tasks via Tool Generation” Proceedings of the Fourteenth International Conference on Learning Representations. (ICLR 2026, CCF-A). [Paper]
• Liu, Mugeng, et al. “A First Look at Bugs in LLM Inference Engines.” ACM Transactions on Software Engineering and Methodology (TOSEM 2026, CCF-A). [Paper] [Repo]
• Liu, Mugeng, et al. “WAB: Overcoming Memory, Network, and Security Walls in Native Agentic Browsers with WebAssembly.” The 1st International Workshop on Foundations and Architectures for the Agentic Web (FAAW @ WWW 2026). (To appear)
• Liu, Mugeng, et al. “WebANNS: Fast and Efficient Approximate Nearest Neighbor Search in Web Browsers.” Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2025, CCF-A). [Paper] [Repo] [Demo]
• Liu, Mugeng, et al. “WebAssembly for Container Runtime: Are We There Yet?” ACM Transactions on Software Engineering and Methodology (TOSEM 2025, CCF-A). [Paper] [Repo]
• Liu, Mugeng, et al. “Research Artifacts in Software Engineering Publications: Status and Trends.” Journal of Systems and Software (JSS 2024, CCF-B). [Paper] [Repo]
• Zhong, Siqi, Liu, Mugeng, et al. “LaTune: Lightweight and Adaptive Configuration Tuning for LLM Inference on Edge Devices.” Proceedings of the ACM Web Conference 2026 (WWW 2026, CCF-A). [Paper]
• Zhang, Yixuan, Liu, Mugeng, et al. “Research on WebAssembly Runtimes: A Survey.” ACM Transactions on Software Engineering and Methodology (TOSEM 2025, CCF-A). [Paper]
• Han, Yudong, Bi, Weichen, Shen, Haiyang, Liu, Mugeng, et al. “Characterizing GPU-accelerated Web Applications in Browsers” Proceedings of the 48th International Conference on Software Engineering (ICSE 2026, CCF-A). [Paper]
• Xie, Yuhang, …, Liu, Mugeng, et al. “From Task to Tutorial: An Automated GUI Framework for Excel Tutorial Document and Video Creation.” Proceedings of the ACM International Conference on the Foundations of Software Engineering (FSE 2026, CCF-A). [paper]
• Shen, Haiyang, Yan, Hang, Xing, Zhongshi, Liu, Mugeng, et al. “DRAGON: Domain-specific Robust Automatic Data Generation for RAG Optimization.” (EACL 2026). [Paper]
• Chen, Qin, Ren, Yuanyi, Ma, Xiaojun, Liu, Mugeng, et al. “SheetDesigner: MLLM-Powered Spreadsheet Layout Generation with Rule-Based and Vision-Based Reflection.” Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025, CCF-B). [Paper]
• Chen, Zhiyang, Ma, Yun, Shen, Haiyang, and Liu, Mugeng. “WeInfer: Unleashing the Power of WebGPU on LLM Inference in Web Browsers.” Proceedings of the ACM Web Conference 2025 (WWW 2025, CCF-A). [Paper] [Repo]

📂 Projects

• ThinkSync AI systems (Dec. 2025 - Present)
  • Leader and core contributor, built a family of real-world LLM agent products, covering web-native automation, personalized AI, reusable skills, and human-agent collaboration.
  • OpenWebClaw (Jan. 2026 - Present)
    • A web-native AI agent developed by our ThinkSync-AI team for personalized assistance and long-running task execution directly on the web.
    • Maintains persistent context, reusable skills, scheduled automation, and site-specific experience so agents can improve from previous tasks rather than starting from scratch.
  • SparkBox (Jan. 2026 - Present)
    • A local-first personalized AI assistant focused on identity continuity, long-term memory, and personal/shared knowledge construction.
  • WebAct (Jan. 2026 - Present)
    • A human-agent tool for constructing reusable web information-collection and crawler skill scripts with typed inputs, guardrails, versioned rollouts, and observable execution.
• ATMOS: AI-Assisted Teaching Mode Optimization System (Mar. 2022 – Feb. 2024)
  • Developed an AI system that automatically analyzes classroom teaching modes and generates optimization recommendations; deployed in Peking University courses.
  • Collaborated with graduate students and faculty across software engineering, artificial intelligence, education, and psychology.
  • Served as the core developer, leading DL-based analysis algorithm design, end-to-end implementation of the cloud-edge-IoT architecture, and system deployment.

💼 Experiences

• Research Intern, Microsoft (May. 2025 - Dec. 2025)
• Undergraduate Student Instructor, EECS, Peking University (Sep, 2021 - Jul, 2025)
  • Instructed over 40 undergraduate students.
• Teaching Assistant, Introduction to Computing (Python), Peking University (Sep. 2021 - Dec. 2025)
  • Instructed over 480 undergraduate students across four semesters.
  • Developed a browser-based Python IDE to provide a zero-installation, privacy-first coding environment for beginners.
  • Developed an LLM-based Cyber TA to offer students automated debugging support.
• Teaching Assistant, Computer Networks, Peking University (Mar. 2022 - Jun. 2022)
  • Instructed over 80 undergraduate students.

🎖 Selected Awards

• Merit Student, Peking University (2025)
• Tencent(腾讯) Scholarship, School of Computer Science, Peking University (2025)
• Award for Scientific Research, Peking University (2024)
• Luo Yuehua Scholarship, Peking University (2024)
• Merit Student, Peking University (2023)
• Ubiquant(九坤) Scholarship, School of Computer Science, Peking University (2023)
• National Scholarship, Ministry of Education (2021)
• Beijing Merit Student, Beijing Municipal Education Commission (2021)
• Merit Student Pacesetter, Peking University (2021)