A Multimodal Adaptive Guided System Based on Large Language Models

The proposed system establishes a multimodal tour guide framework by leveraging large language model (LLM). At the front-end, it integrates multi-source sensory devices to capture real-time visitor data. The back-end, powered by advanced YOLOv8 and EfficientNetV2 models alongside the ChatGLM3 model, runs on domestically developed TPU chip BM1684X, optimized for high efficiency and low power consumption. This architecture facilitates personalized interpretation services tailored to visitors by gender and age classification. Additionally, the system includes interactive functionalities such as speech recognition and a robust knowledge retrieval database, ensuring intelligent management and optimization of tourism services.