Технологія розроблення рекомендаційного чат-бота на основі великих мовних моделей для проєктування ІоТ-системи

Abstract

Modern development of information technologies, particularly the Internet of Things (IoT), opens new opportunities for automating the design of systems operating on low-power wide-area networks (LPWAN). This requires novel approaches to developing the architecture of such systems and selecting technological solutions that account for the specific conditions of their use. The study utilized large language models (LLM) to create a chatbot that provides recommendations. The model was trained on a large dataset collected from scientific articles and technical documents related to LPWAN. Additionally, automated generation of structured materials for system design was implemented, considering the specific features of such networks. Main results include highly detailed and formalized stages of the developed technology, which enables the automatic formation of IoT system architectures and corresponding process flow diagrams (PFDs) based on user-provided textual descriptions. Key LPWAN characteristics and constraints, such as low power consumption, extensive but not unlimited coverage, and limited bandwidth, were taken into account. This is crucial for building efficient and scalable solutions. An approach to adapting the language model to specialized LPWAN terminology and context was developed. Techniques for dataset augmentation were described, leveraging specialized libraries and methods. As a result, the chatbot can provide accurate and professional advice, valuable for technical specialists. For instance, it suggests optimal data transmission methods under constraints on message size or transmission frequency. The quality of responses to complex technical questions has significantly improved compared to baseline model versions. Additionally, automated graphical visualization of data flows in the form of PFDs was implemented, facilitating understanding of system structure and processes. This enhances analysis and decision-making. Future research prospects involve refining the technology by expanding the knowledge base with up-to-date data and adapting it to other IoT protocols to ensure broader applicability and functionality.