My name is Tianqi Liu (zh: 刘天琦). I am currently a second-year MS student in Zhejiang University, under the supervision of Prof.Shengyu Zhang.
My research interests primarily focus on LLM agents and recommender systems, particularly in agent post-training， customization, and realistic evaluation. Recently, I have developed a growing interest in personalized, user-centric, self-evolving LLM agents, seeking to tackle the challenges of proactive interaction in human-agent interaction.
🔥 News
- 2026.07: 🎉🎉 One paper has been accepted to EMNLP 2026 Oral.
- 2026.04: 🎉🎉 One paper has been accepted to ACL 2026.
- 2025.10: 🥳🥳 I went to Dublin, Ireland, to deliver an oral presentation of our paper CHORD in ACM MM 2025.
- 2025.07: 🎉🎉 One first-author paper has been accepted to ACM MM 2025.
📝 Publications
Tianqi Liu, Yandong Guan, Songze Li, Wenkai Wang, Gang Yu, Chengfei Lv
- Enterprise development goes beyond isolated fixes. Coding agents must fulfill evolving business requirements across mature repositories, existing interfaces, and dependent modules.
- Evaluate complete delivery, then explain the gaps. PACE-Bench links unmet PRD requirements to code and execution evidence, helping teams diagnose integration failures and refine coding-agent harnesses.
Tianqi Liu, Kairui Fu, Shengyu Zhang, Wenyan Fan, Zhaocheng Du, Jieming Zhu, Fan Wu, Fei Wu
- A framework for device-cloud collaborative personalized mixed-precision quantization that achieves model customization and compression with one forward pass.
- Frozen weights + Channel-wise quantization strategy = Fast AND Personalized model adaptation.
What’s Missing in Screen-to-Action? Towards a UI-in-the-Loop Paradigm for Multimodal GUI Reasoning
Songze Li, Xiaoke Guo, Tianqi Liu, Biao Yi, Zhaoyan Gong, Zhiqiang Liu, Huajun Chen, Wen Zhang
- A GUI reasoning framework utilizing a ‘Screen-UI Elements-Action’ loop to achieve interpretable decision-making and precise interface parsing.
- UI-Driven RL (localization + semantics + utilization) + 26K UI-Comprehension Benchmark = Transparent multi-step GUI automation.
💡 Projects
Evaluating complete business requirements in mature enterprise repositories, with traceable delivery diagnosis to guide coding-agent improvement.
A benchmark for desktop GUI agents with 538 executable tasks in a live Ubuntu environment, covering professional workflows and human-in-the-loop collaboration.
A UI-in-the-loop paradigm for multimodal GUI reasoning, reframing the screen-to-action process as a cyclic “Screen–UI Elements–Action” loop for GUI agents.
🎖 Honors and Awards
- 2026.10 National Scholarship
- 2025.06 Outstanding Graduates of Tongji University
- 2024.12 Outstanding student of Tongji University (Top 5%)
- 2024.12 Scholarship of Tongji University
- 2023.12 Scholarship of Tongji University
- 2022.12 Outstanding student of Tongji University (Top 5%)
- 2022.12 Scholarship of Tongji University
📖 Educations
- 2025.09 - 2028.06, Master, Artificial Intelligence, Zhejiang University.
- 2021.09 - 2025.06, Undergraduate, Software Engineering, Tongji University.
💻 Internships
- 2026.06 - present, Alibaba Taobao&Tmall Group, China.
- 2025.08 - 2026.05, Huawei Markov Lab, China.
- 2025.03 - 2025.05, Huawei Noah’s Ark Lab, China.