BDI-Driven Indoor Positioning and Assistance via Hermes Mesh Networks and LLM Interfaces

Recent years have seen an overwhelming diffusion of digital assistants thanks to an endless amount of resources being employed in the research and production of sharper, faster and more powerful neural networks. These advancements have enabled users from diverse backgrounds to access complex datasets through simple natural language inquiries. Using only natural language, users can interact with data fetching services without needing to understand their underlying technical implementation. Our project takes root in such an ever-changing environment and addresses the long-standing necessity of an indoor localisation system in medium-to-large buildings, such as offices or schools. We will present the potential of our architecture, achieved via a state-of-the-art implementation of a mesh communication network between ESP32 microcontrollers based on advanced technologies such as LoRa, Bluetooth Low Energy and the framework Hermes. Pursuing maximum adaptability, the whole application has also been modelled over the BDI paradigm, given its superb information hiding capabilities and ease of understanding. To facilitate the process, a specialised custom library called DEMOCLE was fine-tuned and employed, which enables distributed knowledge programming on embedded devices. Finally, user interaction is managed by a Large Language Model-based artificial intelligence, capable of translating human requests into SQL queries that precisely access the collected data to meet the inquirer’s needs.