Tiancheng Xing Yuu
Undergraduate · Artificial Intelligence · Renmin University of China
I’m an undergraduate at the Gaoling School of Artificial Intelligence, Renmin University of China. My research interests lie in multimodal reasoning, efficient multimodal models, and video generation.
I am particularly interested in understanding and reducing computational redundancy in multimodal models through KV cache compression, token pruning, sparse attention, and adaptive caching.
I also have experience in visual SLAM and robotic perception, which has shaped my interest in building efficient, robust systems that can perceive, reason about, and interact with the real world.
Research interests
Efficient Multimodal Reasoning
Investigating how vision–language models retain and use visual evidence during multi-step reasoning, with a focus on visual information flow and efficient inference through KV cache optimization, token pruning, and sparse attention.
Video Generation & Post-Training
Exploring temporally coherent video generation through efficient memory and feature reuse, alongside post-training methods based on reinforcement learning and preference alignment to improve visual quality and prompt adherence.
Selected publication
All publicationsConstraint-Guided Prompting and Semantic-Aware Evaluation for LLM-Based ABSA
ICIC 2026 · Springer LNAI
Constraint-Guided Prompting (CGP) for more reliable structured extraction with large language models, and Sem-F1, a semantic-aware evaluation protocol for Aspect-Based Sentiment Analysis.
BibTeX
@inproceedings{li2026constraint,
title = {Constraint-Guided Prompting and Semantic-Aware Evaluation for LLM-Based ABSA},
author = {Li, Muzhi and Xing, Tiancheng and Wang, Yuheng},
booktitle = {ICIC 2026},
year = {2026},
doi = {10.1007/978-981-92-3435-6_2},
url = {https://link.springer.com/chapter/10.1007/978-981-92-3435-6_2}
}
Selected projects
GitHubNeneBot
Retrieval-augmented generation
A character conversational AI grounded in original game scripts. Combines FAISS semantic retrieval, pluggable local and cloud LLM backends, multi-turn memory, SSE streaming, and a Vue 3 visual-novel interface.
Python · FastAPI · FAISS · Vue 3
SLAMForge
Visual SLAM & dense reconstruction
A C++20 monocular visual SLAM and dense reconstruction system inspired by ORB-SLAM3. Combines geometric tracking, local bundle adjustment, Sim(3) loop closure, and pose-graph optimization with learned depth and multi-view consistency.
C++20 · Python · Ceres · g2o
Recent writing
All articlesNotes and experiments in artificial intelligence and computer systems.