HOLLOW TREE: Harnessing Online Learning with Large-language-model Oriented Web-apps for Teaching, Research, Evaluation, and Engagement

Recent advances in Artificial Intelligence (AI), in particular the applications of large language models (LLMs), have opened new avenues for enhancing educational experiences. One such opportunity is for students to engage with AI to help review concepts taught in class. In this paper, we introduce the HOLLOW TREE project or Harnessing Online Learning with Large-language-model Oriented Web-apps for Teaching, Research, Evaluation, and Engagement. The system comprises two interconnected applications. The first one is SCUIRREL or Science Concept Understanding via Interactive Reinforcing Review by Educational Large-language-model. SCUIRREL is a chatbot that will assist students in reviewing specific course topics though engaging in a guided conversation. The second, instructor-facing, application is called ACCORNS or Admin Control Center Organising the Resources Needed for SCUIRREL. Here instructors can upload relevant course materials and create or manage specific topics that SCUIRREL should review with the students. By providing a set of concepts (i.e. key takeaways) for each topic, the instructor provides SCUIRREL with specific guidance on what aspects of a topic to review with the students. In addition, ACCORNS can assist in generating quiz questions based on the provided topics that after instructor validation can be accessed by students in SCUIRREL as an additional way of assessing content understanding.Leveraging techniques such as prompt engineering and retrieval augmented generation, the system ensures that tutoring interactions remain focused on course-specific objectives while dynamically adapting to the learner’s level of understanding. Preliminary evaluations—including beta testing with experienced educators and a pilot classroom trial—demonstrate the potential of the HOLLOW TREE apps to enhance learning outcomes, promote student engagement, and provide valuable data for assessing both technical performance and educational impact. Our findings underscore the promise of integrating AI-driven tutoring with robust instructor control, while also highlighting challenges related to user adaptation and AI literacy. Future work will focus on refining the system, expanding its applications to various educational contexts, and conducting comprehensive evaluations to further validate its effectiveness as a tool for modern education.