Leveraging Large Language Models for Statistical Analysis of Aggregated Configurations of Physical Intrusion Detection Systems

This study investigates the application of Large Language Models (LLMs) combined with Retrieval- Augmented Generation (RAG) techniques to extract statistical insights from anonymous aggregated configurations of Physical Intrusion Detection Systems (PIDS), commonly deployed in residential and office buildings. Individual PIDS configurations, originally stored in separate databases, were consolidated into a centralized MySQL database. A custom LLM-powered pipeline was developed for processing natural language queries, dynamically generating SQL statements based on the relevant data retrieved, and producing natural language summaries. This approach enables non-technical users to access complex technical setups through conversational interfaces. The resulting insights can support decision-making in various domains, including marketing strategies, new product planning, and prioritization of software and firmware development. This study demonstrates the potential of combining structured data aggregation with advanced natural language processing techniques to automate high-level reporting from complex datasets.