The aim of the project is to develop an LLM-based classifier of water distribution system status. The proposed project builds on the research work of the WaterPrime project in which an advisory system was developed for condition diagnosis of District Metered Areas (DMAs) including leak detection and alarm events. The system was a multimodal time series classifier (i.e. from different sources and with different data types, TS) based on random forests. In addition to its high performance, its important feature was its explainability - the ability to present the user with logical reasoning to justify the state label in the form of logical rules, referring to DMA parameters. The capabilities of this classifier represent the state of the art to date for advisory systems using classical ML models. The intensive development of LLM models over the past year provides the opportunity to develop a new advisory system with a high-performance  classifier capable of interacting with an expert and justifying decisions in natural language. The outcome of the project will be a classifier algorithm including a data transformation method, construction and learning of the algorithm, and evaluation of its effectiveness of performance.