Weile Guo

I am a second-year master student in the School of Computer Science and Technology at Sun Yat-sen University, advised by Prof. Chao Yu. I received my B.Eng. degree from Huazhong University of Science and Technology (HUST) in 2023.

Beyond research, I enjoy building practical projects and exploring AI-driven development workflows. In my free time, I enjoy cycling, watching movies, and exploring good food.

Email / CV

Work Experience

AI Application Engineer Intern Huawei July 2026 – Present

Core Technology Researcher Intern iFLYTEK July 2021 – September 2021

Publications

MotionHalluc: Diagnosing Kinematic Hallucinations in Fine-Grained Motion Reasoning

Weile Guo, Shenghong He, Danying Mo, Chengdong Xu, Xuexun Liu, Chao Yu

NeurIPS 2026 ED Track — Poster

[arXiv]

Projects

My research interests lie in multimodal large language models, computer vision, and multimodal reasoning. Currently, I focus on fine-tuning and enhancing multimodal large models for motion understanding and reasoning tasks, with particular interests in hallucination mitigation, physics-aware reasoning, and vision-language alignment. I am also interested in building reliable multimodal AI systems that bridge visual perception and structured physical knowledge for real-world applications.

	AI-assisted F1 Content Operations Pipeline Weile Guo. Personal project, 2026.06 Xiaohongshu link Built an AI-assisted F1 Chinese content operations pipeline from product requirements to an end-to-end MVP. I analyzed the content workflow and split it into two production paths: an automated OpenF1 data line for race and session reports, and a keyword-driven Agent line for controversy tracking and topic exploration. The system combines FastAPI, Postgres/pgvector, APScheduler, a Vite React review console, Playwright-based rendering, and Feishu notifications to turn structured data and editorial intent into reviewable social media drafts.
	Motion Correction Workflow with Agent-Verified Skeletal Metrics Weile Guo. On-campus project, 2026.01 Developed an agent-based sports coaching system that combines multimodal video comparison, DTW-based motion alignment, and tool-augmented reasoning to generate explainable motion correction feedback. The Agent autonomously verifies uncertain predictions using skeletal measurement tools and evidence-based reflection, improving correction accuracy by 5% while reducing false correction rates by up to 47.5% across evaluated VLMs.
	AI-based Motion Diagnosis and Visual Reporting System Weile Guo. On-campus project, 2025.08 This project aims to automatically compare student movements with expert demonstrations and generate actionable correction feedback. I developed an end-to-end pipeline that performs pose sequence extraction, temporal alignment, key-frame analysis, textual correction generation, and visual report rendering. The system supports fast review of motion differences and provides a clear, interpretable output for teaching and self-correction.
	Vision-based UE Interactive Fitness Game Weile Guo. Undergraduate thesis, 2023.05 Watch Demo Video This project explores camera-based human-computer interaction for fitness gaming without wearable devices. I independently designed a modular system that links pose recognition with the game engine, enabling real-time motion input through body keypoints extracted from video. The prototype supports multiple interaction modes and game features, demonstrating the feasibility of vision-driven interactive gaming with stable real-time performance.

Thanks to Jon Barron for open-sourcing the page, see source code.
Last updated: