AI Agents
Planning, tool use, multi-agent collaboration, memory, reliability, and long-running agent workflows.
AI researcher · builder · founder
Shaoxu Sun
Ph.D. student at the USTC Lab for Data Science, jointly trained with Zhongguancun Academy in Beijing.
I work on AI agents, token-efficient AI systems, and the economics of foundation-model services—with a strong interest in turning research ideas into useful products.
Researching agents. Building AI products.
About
Research with a builder’s mindset.
I am interested in how intelligent systems can plan, collaborate, use tools, and allocate model resources under real cost and performance constraints. My current work sits at the intersection of agentic AI, efficient foundation-model systems, and model-service economics.
Outside research, I enjoy building and shipping products. I founded TokenLab, and I am especially drawn to entrepreneurship where infrastructure, AI capability, and user experience meet.
Research focus
What I think about.
Token-efficient AI Systems
Token budgeting, accounting, caching, context management, and system design that improves capability per unit cost.
Model & Resource Routing
Adaptive selection across heterogeneous models and resources under quality, latency, and cost constraints.
Broader Interests
Multimodal intelligence, reinforcement learning, retrieval-augmented systems, and scalable AI services.
Selected work
Research, infrastructure, and products.
Founder
tokenlab.pro ↗
TokenLab
AI model access and infrastructure for developers, built around practical model usage, token-aware metering, routing, and reliable developer workflows.
TokenLab reflects my interest in the real economics of foundation models: how tokens are measured, priced, routed, and turned into dependable AI products.
Explore TokenLab ↗CiwardClaw
A personal OpenClaw fork focused on agent reliability, provider compatibility, routing, recovery, and cost guardrails across real-world workflows.
View project ↗DMR4Rec
Dynamic model routing for sequential recommendation, exploring instance-level routing across heterogeneous recommendation architectures.
Research projectCampus Q&A
A retrieval-augmented question-answering system for campus knowledge, combining practical retrieval, generation, and product engineering.
View project ↗Education
Background.
Current
University of Science and Technology of China
Ph.D. student · USTC Lab for Data Science
Joint training program with Zhongguancun Academy, Beijing
2021–2025
Shandong University
B.Eng. in Artificial Intelligence
School of Computer Science and Technology
Writing
Notes from building and learning.
The original blog is preserved as an archive of technical notes, course projects, experiments, and things I learned along the way.
Contact
Research, products, or something worth building?
Feel free to reach out.