Gen-Presenter: An Interactive AIoT Presentation System with RAG-Supported LLMs

Gen-Presenter is a cost-effective, AI-powered interactive presentation system designed for consumer-facing environments such as museums, exhibitions, and public installations. Addressing the limitations of traditional kiosks and robotic guides, it leverages Retrieval-Augmented Generation (RAG), large language models (LLMs), and edge AI to deliver adaptive multimodal content tailored to visitor demographics and engagement levels. Running primarily on an edge device, the system integrates modules for gaze detection, age estimation, automated speech recognition (ASR), text-to-speech (TTS), and content management. Two LLMs on a backend NLP server handle slide selection and explanation generation. While LLM2 with RAG enriched responses with contextual content, LLM1 (LLaMA 3.3 without RAG) demonstrated superior slide selection performance, achieving 0.95, 0.91, and 0.96 accuracy with only 1s of latency. User suitability testing demonstrates strong alignment with diverse age groups (scores between 0.81 to 0.96), and the multithreaded design delivers a 15.7x improvement in slide processing efficiency. These results highlight the system’s scalability, responsiveness, and potential as an intelligent consumer electronics solution for immersive public engagement.