Portfolio

Human-Agent Collaboration

I study human-agent collaboration along three complementary lines: ALMANAC captures how humans form and update these mental models during collaboration; CollabSim investigates LLM agents' collaborative competence through controlled multi-agent experiments; our Human-Agent Collaboration Platform provides an open-source research platform for HCI researchers studying human–LLM agent collaboration.

Preprint

ALMANAC

Humans' ALMANAC: A Human Collaboration Dataset of Action-Level Mental Model Annotations for Agent Collaboration

A dataset of humans' Action-Level Mental model ANnotations for Agent Collaboration, built from the Map Task. At every action, participants annotated their mental models: team goals, partner intentions, and their own reasoning.

2,987 Annotated Actions
50 Participants
25 Sessions
Mental Models Common Ground Dataset
ALMANAC action-level mental model annotations
Preprint

CollabSim

A CSCW-Grounded Methodology for Investigating Collaborative Competence of LLM Agents through Controlled Multi-Agent Experiments

A configurable simulation framework grounded in CSCW research. Researchers can systematically vary task paradigms, interaction conditions, model backbones, and agent designs to isolate what drives or breaks collaborative competence in LLM agents.

4 CSCW Paradigms
3+ Interaction Conditions
12+ Model & Agent Designs
Multi-Agent Systems CSCW Simulation
CollabSim experiment controller and agent action flow
CHI 2026

Human-Agent Collaboration Platform

Through the Lens of Human-Human Collaboration: A Configurable Research Platform for Exploring Human-Agent Collaboration

An open-source, extensible research platform for HCI researchers. Its modular design supports adaptation of classic CSCW experiments and theory-grounded manipulation of interaction variables, enabling systematic study of human–LLM agent collaboration.

Customizable CSCW Paradigms
Configurable Experiment Conditions
Open Source & Extensible
Human-Agent Collaboration CSCW Customizable
Human-Agent Collaboration Platform architecture

Education Technology

Designing child-centered AI systems that support personalized learning experiences, grounded in multi-stakeholder perspectives from children, parents, and educators.

CHI 2025

StoryMate

Characterizing LLM-Empowered Personalized Story-Reading and Interaction for Children: Insights from Multi-Stakeholder Perspectives

An LLM-powered personalized story-reading system for children, designed from multi-stakeholder perspectives. StoryMate adapts narrative content and interaction to each child's interests, supporting engaging and developmentally appropriate reading experiences.

Child-Centered AI Education LLM Applications
StoryMate interactive story-reading demo