Designing an Embodied Chatbot for High School Students: An Interaction Analysis

Abstract

This paper presents DragonBot, an embodied retrieval augmented generation chatbot designed by high school students for use on their residential campus. The system combines a large language model with both static and dynamic institutional data to answer questions through a voice-based interface represented by an animated dragon mascot. DragonBot was deployed in a public student area for four days, during which 882 naturally occurring student questions were collected. Using a multistage human and LLM assisted coding process, the authors identified sixteen categories that reflect five broader themes: social interaction, campus life, requirements and rules, attempts to challenge the system, and miscellaneous queries. Response times and answer quality were also evaluated across static, dynamic, and model only inputs, showing that most answers were rated as helpful and that dynamic data produced the highest accuracy. The findings illustrate how high school students interact with an embodied agent in an authentic setting and provide initial evidence about the types of information they seek from such systems. This work contributes one of the first analyses of real world conversational data from a high school chatbot deployment.