Intelligent Sports Engineering @ Zhejiang University
Computer Vision · LLM Agents · Sports Analytics · Computational Forensics

Hangzhou, China · 3240100798@zju.edu.cn

I build end-to-end systems that turn messy real-world inputs — case files, match videos, clinical footage — into structured, auditable outputs. My work sits where computer vision and LLM agents meet: vision measures the physical world, language models organise the evidence, and deterministic rules own every conclusion that matters.

我关注计算机视觉与大模型智能体的结合，把案卷、比赛视频、检验影像这类非结构化输入，做成可复核、可追溯的结构化结论。

• Computational forensics — core member of the Zhejiang University Forensic Identification Center AI project, building an end-to-end joint range-of-motion assessment and report-generation system (since 2025.8)
• Sports knowledge graphs — knowledge graph and QA system for table-tennis coaching, connected to the ZJU Institute of Sport Science and Health Engineering database (2025.4 – 2026.4)
• Sports video understanding — tennis ball, court and bounce tracking from broadcast footage

A multi-agent research prototype for forensic clinical disability assessment. Local-first OCR/LLM extraction, human-confirmed goniometer readings, and a deterministic rule engine for ROM loss, disability clauses and recovery periods.
Python PaddleOCR LLM agents Rule engine

A one-click Streamlit benchmark for comparing LLM accuracy on a table-tennis question set, with connectivity probes, retry/backoff and dual accuracy metrics.
Python Streamlit LLM evaluation

The earlier MMPose + RAG prototype behind ForenAgent: a machine-vision SOP for joint measurement, with 97.11% joint-tracking accuracy at 2.41° mean angular error.
MMPose RAG Computer vision

• Email: 3240100798@zju.edu.cn
• Zhejiang University, Zijingang Campus, Hangzhou
• Open to internship opportunities in computer vision and AI algorithm engineering