Research Projects

The proposed Virtual Teaching Assistant integrates Retrieval-Augmented Generation (RAG) with a Multi-strategy AI Agent to provide adaptive pedagogical guidance. By identifying different question types, the system applies Bloom’s Taxonomy to clarification questions and Guided Dialogue to exploratory questions, offering structured explanations or scaffolding hints that encourage students to construct knowledge rather than simply receive direct answers. 

To promote proactive learning, the system introduces “The Handshake,” a two-level activity combining concept verification with extended challenges. This mechanism encourages students to examine their understanding and apply knowledge in new contexts while lowering barriers to asking questions. A field experiment will analyze learning behaviors and motivation, evaluating whether proactive AI interaction can foster deeper cognitive engagement. 

The 'Assistant' component integrates Retrieval-Augmented Generation (RAG) with AI Agent technology to overcome the limitations of immediate feedback in traditional classroom settings. Through dynamic decision-making, the system autonomously identifies query types and selects optimal tools to deliver curriculum-based Q&A and content summaries. This approach significantly enhances interactivity, establishing a more efficient and responsive environment for personalized guidance.

This component focuses on visualizing the student's cognitive state through real-time feedback and behavioral analysis. Unlike conventional dashboards that merely display completion statistics, the system integrates error distribution, performance metrics, and accuracy-completion contrasts to transform data into motivation for self-reflection. By intuitively presenting these learning trajectories, the portfolio enables students to accurately pinpoint weak areas and optimize the effectiveness of their self-directed learning.

This project aims to develop a personalized intelligent tutoring system through a large language model, encompassing two components: automatic question generation, grading, and personalized feedback. Unlike most intelligent tutoring systems which interact with users in a preset manner, this system focuses on individualized feedback based on each student's performance. The system strives to gain a deeper understanding of students' learning conditions, aiding them in adjusting their learning pace and providing more targeted guidance.