Hello！ I am a Master's Student at Tsinghua University, GIX program (Dual Degree, MS. & Mdes, HCI), also affiliated with the Future Laboratory. I hold a B.E. degree in industrial design from Tongji University.
My technical skills mainly involve generative AI, mechatronics, and design. I am interested in multimodal AI agents with digital or embodied entities that enhance the productivity, cognition, and well-being of humans.
I have related publications at ACM CHI, ACM UIST, IEEE VR, IEEE IROS, ACM TEI, HRI...
Previously, I have interned at the tech teams of Microsoft Research, Moonshot.ai, and Tsinghua University, working on LLM, agentic systems, HCI, and EE.... Moreover, I was a founder of a startup invested by MiraclePlus (YC China) in AI hardware.
Long-Term Research Question：
How can we develop multimodal AI agents with digital or embodied entities that assist humans with life, work, emotion, and self-actualization through natural user interfaces and intent sensitivity?
Peizhong Gao, Ao Xie, Shaoguang Mao, Wenshan Wu, Yan Xia, Haipeng Mi, Furu Wei
We introduce Meta-Reasoning Prompting (MRP), an innovative method for large language models (LLMs) that enhances performance by dynamically selecting reasoning strategies based on task requirements. MRP operates in two phases: identifying the best approach and applying it to the task. Our benchmarks show that MRP achieves or nears state-of-the-art results across diverse tasks, significantly improving LLMs' ability to tackle complex problems efficiently and adaptively.
Yu Zhang, Peizhong Gao, Fangzhou Kang, Jiaxiang Li, Jiacheng Liu, Qi Lu, Yingqing Xu
We present OdorAgent, a system that automates video-odor matching by combining a large language model with a text-image model. This framework operates across four dimensions: subject matter, emotion, space, and time. User studies on a specific movie demonstrated that OdorAgent allows even inexperienced users to create olfactory experiences, showing significant adaptability to different scenes and enhancing viewer engagement and immersion.
Ye Jin, et al.
IROS 24' paper 丨@THU丨Paper 丨 Bibtex
This paper introduces a framework for creating human-like generative driving agents by using post-driving self-reports from drivers as demonstration and feedback. Urban driving experiments captured verbalized thoughts, which served as prompts for an LLM-Agent. Evaluations showed that this approach reduced collision rates by 81.04% and enhanced human likeness by 50% compared to baseline LLM-based agents, highlighting the effectiveness of incorporating expert demonstration data.
Enhancing the Body Language of Robots Responding to Affective Context through LLM
Peizhong Gao, Ziheng Xiao, Ao Xie, Yixin Li
On-going 丨@THU
We have developed a language model-based action generator that enables companion robots to respond to human emotions with rich and contextually appropriate body language. Our initial experiments were conducted in Unity, and the next step will involve considering more stringent physical constraints.
OZ-Dev: A Visual Programming Tool for LLM Apllications with Open-source Hardware
Peizhong Gao, Guanbo Wang, Xiao Yi, Honglin Lyu, Qi Shan, Long Lin, Yuying Jiang
for internal use丨@THU
We have developed a language model-based action generator that enables companion robots to respond to human emotions with rich and contextually appropriate body language. Our initial experiments were conducted in Unity, and the next step will involve considering more stringent physical constraints.
Peizhong Gao, Fan Liu, Di Wen ... Qi Lu, Haipeng Mi, Yingqing Xu
We introduce Mul-O, a task-oriented platform designed to help semi-professionals prototype olfactory experiences across diverse contexts. Mul-O integrates a web-based UI, an API for third-party integration, and wireless olfactory hardware, streamlining multisensory design and concept validation. A 15-day workshop with 30 participants resulted in seven innovative projects, demonstrating Mul-O's effectiveness in advancing olfactory innovation.
Research and Development of A Controllable Olfactory Interface Mobile Terminal
Peizhong Gao
Bachelor's Degree of Engineering Thesis丨@Tongji U
The author developed a compact, controllable olfactory interface mobile terminal, along with its online management software. It integrates a kind of mini air pump within compact mechanical and electronic structures, functioning as a portable IoT scent player, capable of rapid, residue-free scent switching and supporting further development. The thesis provides a detailed description of its principle, design, engineering, and experiment.
CHI 23' poster丨@THU
This paper introduces Atmospheror, an interactive system that utilizes ambient olfactory displays to enhance interaction and feedback in synchronous online classes. A pilot study with two instructors and 13 students assessed its effectiveness, revealing that Atmospheror improves student concentration and interactivity. User interviews provided insights for future enhancements in odor selection, hardware implementation, and interface mechanisms.
Peizhong Gao, Tanhao Gao, Yanbin Yang, Zhenyuan Liu, Jianyu Shi, Jin Kevin Li
This paper explores Digital Craft as the hybridization of digital technologies with human skills. Focusing on bamboo craft, it uses design research methods to reveal innovative digital interventions that enhance the bamboo-making process and integrate 3D printing. The authors present digital toolkits for bamboo weaving and a platform to improve compatibility between computational design and craft.
Percuino: Swapping Senses with Animals by Modular PCB Product
Personal丨@Fablab丨Exhibited at WDCC
SmartHand: Active hemiplegia intelligent rehabilitation system facilitated by data loop of Electromyographic signal and Electrical stimulation
Team丨@Tongjia U
Ani: Conversational Agents with Generative AI for children
Team丨@THU
A Future Marine Vehicle Design Concept for Extreme Sport
Team丨@Tongji U
CareU: Transfer Chair for the Disabled Elderly
Team Project丨A' Design Award 2022丨@Tongji U
Runing Food: A Serious Game for Popularizing the Food Carbon Footprint
Team丨@Tongji U @Tencent
LAMO: A Wearable Device of Lamazer Breathing Training for Pregnant women
Team丨@Tongji
Galley As a Service: Modular Galley Product Service System Design
Team丨@Boeing
Craftsman Jounery: Worker-centered Digital Management Platform
Team丨@BOSCH
Asia, Natural Language Computing Group
LLM Research Intern
2024
Supervisor: Shaoguang Mao, Yan Xia
# Paper [1st author, submitted]: “Meta Reasoning for Large Language Models”
- Propose a dynamic reasoning method MRP inspired by human thinking pathway, improving LLMs performance on comprehensive benchmarks with nearly 10%, reaching SOTA.
- Implement experiments and analyze the results.
- familiar with various LLM engineering tech, such as Prompt Engineering, Fine-tuning, RAG, and Agent.
# Product Manger: ChatGPT-like web product for kosmos and bitnet (Stealth)
- Serve as a product manager in function definition, interactive prototyping, data visualization, testing. The product reports to Bill Gates.
- Responsible for organizing weekly technical seminars of LLMs, collecting and sharing papers on arxiv, huggingface, etc.
Future Laboratory, Tsinghua University
Milab & Olfactory Computing Group
Graduate Research Assistant
2022-2024
Supervisor: Haipeng Mi, Qi Lu, Yingqing Xu
# Paper [2nd author, IEEE VR]: “OdorAgent: Generate Odor Sequences for Movies Based on Large Language Model”
- Propose a v-LLM agent system for mapping images and smells in VR.
- Design ablation experiments and user studies that are praised by reviewers.
# Paper [1st author, ACM UIST]: “Mul-O: Encouraging Olfactory Innovation in Various Scenarios Through a Task-Oriented Development Platform”
- Full-stack development of a IoT device for creative support in olfactory interface, including circuits and PCB, mechanical structure, programming, web-APP design.
- Hold a hackathon for 30 people to test our creative toolkits and platform, and conduct qualitative analysis.
- In a related project, responsible for quantitative analysis.
China
Algorithm Intern, Alignment Team
2024-2025
# OS Agent
- Grounding, Benchmark, Mid-training, RL, Evaluation...
# Audio
- Router model training
- Pipeline for SFT, DPO..
- Engineering for LLM
2024
Product Intern, Product Team
# Audio
- prompt engineering
- data pipeline for auto-eval, cleaning, pre-roll...
- large-small model collaboration
# Memory
- Memory mechanism for Kimi
China, Designaffairs Group
Research and Strategy Intern
2022
- User study & data analysis: in POC project for Honda and Stanley, help with preparing 4 focus group (stimulus preparation, observation, recording), and 14 in-depth interview. Analyse the recording for quotes and insights.
- Product defination & design: help with HMI design process (ideation, user study, sitemap, wireframe, high-fi UI...)
Seattle
Contract, University-Industry Cooperation
2022-2023
# Intelligent aviation horse stable system
- Prototyping: Help build IoT devices and information management software for sensing horses' physical status. And test them in the stable.
- Design: help with interaction design of UI for monitoring onboard.
# GAAS - Modular Galley Service Design
China
Contract, University-Industry Cooperation
2022
# Worker-centered”digital management platform
- Ethnography: Explore the living conditions and occupational needs of construction workers and write reports
- Solution: propose main idea and lead team to design an APP in helping workers achieve better career
- The scheme was approved by Bosch
Founder, CEO (Company suspension)
2024.01-2024.07
OZ is dedicated to AI systems with Cloud-End collaboration, applied to embodied intelligence, personal devices, and smart homes. It aims to promote low-cost, high-efficiency model inference, bringing the vision of intelligent connectivity to life.
Entrepreneurship
Selected Honors
China International College Student Innovation Competition
[National Gold Prize]
Peizhong Gao, et al.
News
Baidu “Ernie Bot” Venture Contest [Second Prize]
Peizhong Gao, et al.
Professional Service
ACM CHI Conference on Human Factors in Computing Systems (CHI) 2025
ACM Conference on Intelligent User Interfaces (IUI) 2025
ACM International Conference on Tangible, Embedded and Embodied Interaction (TEI) 2025
ACM Conference on Designing Interactive Systems (DIS) 2024
ACM CHI Conference on Human Factors in Computing Systems (CHI) 2024
Vice President, THU MakerSpace, Tsinghua University
President, Student Arts Society, Tongji University
Board Member, Tsinghua University Gold Award Club