Research

My research have two main focuses: achieveing situational awareness in AI through multimodal learning, including 3D/4D reconstruction, depth sensing, and fine-grained skill analysis.
Then, I tried to make AI trustworthy and efficient under deployment constraints: I have worked on bias evaluation for LLM compression, deployment optimization on low-power hardware, and abalation tests for multi-agent.
Lastly, I'd like to push the application of AI into healthcare, things I've attempted includes medical image analysis, surgical video coaching, and brain-computer interface.

Selected Publications

[1] AutoResearchClaw: Self-Reinforcing Autonomous Research with Human-AI Collaboration

Jiaqi Liu*, Shi Qiu*, Mairui Li, Bingzhou Li, Haonian Ji, Siwei Han, Xinyu Ye, Peng Xia, Zihan Dong, Meng Chen, Congyu Zhang, Letian Zhang, Guiming Chen, Haoqin Tu, Xinyu Yang, Lu Feng, Xujiang Zhao, Haifeng Chen, Jiawei Zhou, Xiao Wang, Weitong Zhang, Hongtu Zhu, Yun Li, Jieru Mei, Hongliang Fei, Jiaheng Zhang, Linjie Li, Linjun Zhang, Yuyin Zhou, Sheng Wang, Caiming Xiong, James Zou, Zeyu Zheng, Cihang Xie, Mingyu Ding, Huaxiu Yao  (*equal contribution)

Under review, ICLR, 2027

[2] The Asymmetric Harms of LLM Compression

Yuan Wu*, Mairui Li*, Lesia Semenova, Chudi Zhong  (*equal contribution)

Preprint, 2026

[3] Test-Time Adaptation for Non-stationary Time Series: From Synthetic Regime Shifts to Financial Markets

Yurui Wu*, Qingying Deng*, Wonou Chung*, Mairui Li*  (*equal contribution)

Preprint, 2025

Research Experience

Surgical Coaching AI

First Author · Cross-functional team; Supervisor: Gedas Bertasius

May 2026 – Aug. 2026

Medical AI LLM/VLM AI-Human Verification Expert Evaluation

Led cross-functional teams to design a medical AI coach that identifies and corrects surgical errors, generating 2B Q&A pairs. Architected an AI-human verification pipeline integrating two automated LLM/VLM checkers with human medical expert evaluators.

SpiderCam Extension for Depth Imaging

Co-first Author; Supervisor: Emma Alexander

The Asymmetric Harms of LLM Compression

First Author · Advisor: Chudi Zhong, Ph.D.

UNC Chapel Hill · Mar. 2025 – Present

VLMs LLMs Pruning Quantization Fairness

Leading evaluation of compressed large language models, focusing on reliability, calibration, and deployment tradeoffs. Developing benchmarking pipelines across pruning and quantization methods to analyze performance and robustness under constraints.

Embedded RL Systems for Adaptive Neurostimulation (Honors Thesis)

First Author · Advisors: Samarjit Chakraborty, Ph.D.; Raghav Pothukuchi, Ph.D.

UNC Chapel Hill · Aug. 2024 – Apr. 2025

Reinforcement Learning Embedded Systems INT8 Quantization ESP32-S3

Designed a reinforcement-learning agent for adaptive embedded deep brain stimulation, reducing Parkinsonian biomarker activity by 7%. Performed FP32-to-INT8 quantization and benchmarked latency, power, and inference for low-power deployment.

AutoResearchClaw: Self-Evolving Autonomous Research Multi-Agent System

Core Researcher · Advisors: Huaxiu Yao, Ph.D.; Mingyu Ding, Ph.D.

UNC Chapel Hill · Feb. 2025 – Apr. 2025

Agentic AI Ablation Studies Human-in-the-Loop

Developed an 82-task, multi-seed evaluation suite enabling multi-agent peer review and reducing data-extraction and API-call costs through a bounded REFINE–REDESIGN–PIVOT policy that reruns failed experiments at minimal cost.

3D/4D Multimodal Neuroimaging Data Pipeline

Core Researcher · Advisor: Margaret Sheridan, Ph.D.

UNC Chapel Hill · Oct. 2024 – May 2025

Neuroimaging fMRI Multimodal Dimensionality Reduction

Conducted multimodal preprocessing across 30+ subjects and 5 conditions, reducing batch runtime by 21%. Modeled spatiotemporal response patterns with region features and dimensionality reduction.

Deep Learning for Non-Stationary Time Series

Core Researcher · Advisor: Yao Li, Ph.D.

UNC Chapel Hill · Jan. 2025 – Dec. 2025

Time Series Test-Time Adaptation Distribution Shift

Studied test-time adaptation for non-stationary forecasting under synthetic drift and financial regime shifts.

Structured Prompting for Human-LLM Interaction Study

Core Researcher · Advisor: Peter Gordon, Ph.D.

UNC Chapel Hill · Dec. 2023 – May 2024

LLM Prompt Engineering NLP

Built output validation and constrained response schemas, reducing malformed or off-task outputs by 60% in pilot testing. Added run-level traceability for a 500-participant multi-turn study.

Teaching & Service

President — MAPS Research Non-Profit

Leadership

Leading a research non-profit organization that aims to fill the lack of Chinese-specific data in mental health datasets and research.

Teaching Assistant & Project Manager

COMP 301 Object-Oriented Programming, UNC Computer Science

Mentored 120 students across 30+ software projects through architecture reviews, Git workflows, and sprint planning.

President — Stand-up Comedy Club

Leadership · Live Performance

Managed 110 student comedy live shows, mentored new performers, and facilitated audience-focused crowd work.