undergraduate student
Tianjin China
shixinyuan20040121@gmail.com
Skills
python
pytorch
linux
git
Languages
Chinese
English
I'm a final-year undergraduate student at Tianjin University of Commerce, focusing on research in computer vision and large-scale multimodal models.
From Sep 2023 to May 2024, I was a research intern at Megvii Research, supervised by Yuyang Liu
From Apr 2025 to Sep 2025, I was a self-funded visiting student of Mohamed Bin Zayed University of Artificial Intelligence, working with researcher Meng Cao and Professor Xiaodan Liang
Engineered advanced AI algorithms and advanced applications at Megvii Research.
Contributed to building the AI Service (AIS) infrastructure, enabling mass algorithm deployment via a zero-code, automated productivity platform.
Introduced the flexible docking task to jointly predict ligand and pocket sidechain conformations for more realistic induced-fit modeling.
Developed a diffusion-bridge generative model over geometric manifolds with energy-to-geometry mapping inspired by Newton–Euler mechanics.
Proposed COLT, a continual tool-usage framework for video LLMs that leverages a learnable tool codebook to dynamically acquire new tools without catastrophic forgetting.
Constructed VideoToolBench, the first video-centric tool-use instruction-tuning dataset, enabling robust evaluation of tool invocation and lifelong learning.
Achieved state-of-the-art results on MSRVTT-QA (+8.2% accuracy over prior SOTA) and substantial continual learning gains on VideoToolBench.
* 1st Prize Academic Scholarship, Tianjin University Of Commerce
* M Award in Mathematical Contest in Modeling
* First Prize in Tianjin Region, China Undergraduate Mathematical Contest in Modeling
I have built a strong presence as a dance content creator on TikTok and Kuaishou, with around 200,000 followers and a cumulative 20 million+ views. My engaging and creative dance videos have attracted a wide audience and established me as a recognized dance influencer on both platforms.