Harry Ye叶 润龙

We design systems that prioritize human agency during qualitative analysis. ScholarMate introduces a mixed-initiative spatial canvas for information sensemaking, while Reflexis operationalizes Design for Deliberation to support rigor in collaborative thematic analysis. Collectively, they illustrate a paradigm where AI scaffolds deep interpretation and reflexivity rather than automating them away.

ARC is an open-source tool designed to streamline systematic literature reviews in computing research. Developed through user-centered design with 20 experienced researchers, it automates literature searches, data extraction, and reference tracking while maintaining transparency and reproducibility standards.

We introduce pedagogical prompting: a theoretically-grounded approach to transform how students interact with AI. We built an interactive AI-system using scenario-based instruction to train this skill, enabling students to leverage LLMs for active learning.

Developing CopilotLens, a framework that transforms AI code generation from opaque suggestions into transparent, explainable interactions. By providing post-hoc summaries and on-demand explanations, we aim to foster more trustworthy human-AI collaboration in software development.

CodeAid is an LLM-based AI programming assistant deployed in a university classroom of 700 students. Through analysis of 8,000 usages, student surveys, and educator feedback, we investigated how AI programming tools can balance educational value with practical support.

OnTrack investigates how email reminders affect student learning behaviors through randomized A/B testing. Using multi-armed bandit algorithms, we optimize intervention strategies to support student engagement while measuring behavioral impacts on academic performance.