Xianyang Liu
✉️ liuxianyang98@gmail.com
I am currently a Research Assistant (remote) at UC Berkeley, working with Postdoctoral Researcher Shangding Gu and Prof. Dawn Song. Previously, I was a Research Assistant at The Hong Kong University of Science and Technology (Guangzhou), advised by Prof. Jiaheng Wei. Before that, I was a Machine Learning Engineer at Ant Group from July 2024 to July 2025. I received my Master’s degree from King’s College London.
My research interests focus on developing and benchmarking Multi-Agent LLM/VLM Systems to enhance reasoning capabilities and enable complex real-world interactions:
-
Evaluation & Benchmark: exploring multi-agent LLM/VLM systems for buyer-seller transaction negotiations [AgenticPay]. 📍UC Berkeley
-
Post-training & Reasoning: generating synthetic reasoning data via multi-agent frameworks to enhance mathematical reasoning [AgenticMath]. 📍HKUST(GZ)
-
Multi-Agent Framework: developing knowledge graph-driven agent orchestration framework that combines LLM/VLM with Eventic Knowledge Graph (EKG) for complex reasoning and online collaboration [MuAgent
]. 📍Ant Group
news
| Mar 02, 2026 | 🎉 Our AgenticMath paper is accepted by ICLR 2026 DATA-FM Workshop! |
|---|---|
| Mar 01, 2026 | 🎉 Our AgenticPay paper is accepted by ICLR 2026 AIMS Workshop! |
| Feb 06, 2026 | 🎉 Check out our new paper in arxiv: AgenticPay: A Multi-Agent LLM Negotiation System for Buyer–Seller Transactions. |
| Oct 22, 2025 | 🎉 Check out our new paper in arxiv: AgenticMath: Enhancing LLM Reasoning via Agentic-based Math Data Generation. |
| Jul 31, 2025 | Jointed The Hong Kong University of Science and Technology (Guangzhou) as Research Assistant. |
selected publications
- ICLR 2026 workshop
AgenticMath: Enhancing LLM Reasoning via Agentic-based Math Data GenerationICLR 2026 Workshops on Navigating and Addressing Data Problems for Foundation Models. [Paper] , 2025