A Multimedia Interactive Presentation System Based on AIoT and RAG-Enabled Large Language Models
Mingshun Wang, Ming-Che Chen · 2025
This paper proposes a multimedia interactive presentation system, Gen-Presenter, that leverages Artificial Intelligence of Things (AIoT) and Retrieval Augmented Generation (RAG)-enabled Large Language Models (LLM). The Gen-Presenter system integrates an edge-AI computing device with peripherals such as a camera, microphone, speakers, and display screen, in combination with a natural language processing (NLP) server. The system detects visitor activity through the camera, infers their age group, and uses this information to select appropriate slides and generate corresponding voice narration for interactive presentations. Experimental results show that Gen-Presenter, when responding to queries from users of different age groups, can select appropriate presentation slides and generating corresponding explanations, with performance closely matching human decisions. In terms of slide selection, the overall precision, recall, and accuracy were 0.87, 0.67, and 0.76, respectively. Additionally, the suitability rates for the generated age-appropriate text content across the three age groups exceeded 0.7. This demonstrates the system's success in delivering human-like slide explanations and interactive Q&A sessions.