A Cloud-Based Smart Water Monitoring System with Integrated Conversational AI: Implementation and Case Study in the Piedmont Region

This thesis presents the design, development, and implementation of an intelligent cloud-based infrastructure for smart water distribution monitoring, featuring an innovative conversational AI interface that transforms how operators interact with complex monitoring systems. The research addresses key limitations in current water infrastructure monitoring, where traditional interfaces require technical expertise and hinder effective system use. The implemented solution leverages Google Cloud Platform services to build a comprehensive monitoring ecosystem integrating real-time sensor data processing, hydraulic simulation, machine learning-based leakage detection, and a conversational AI interface. The architecture applies cloud-native principles such as microservices, serverless computing, and event-driven processing to ensure scalability, reliability, and cost-efficiency. A core innovation is the development of a conversational AI system using Large Language Models and the LangChain framework, enabling natural language interaction with monitoring dashboards. The technical implementation includes automated data ingestion via Cloud Functions, BigQuery for data warehousing, Grafana-based dashboards with geospatial mapping, EPANET integration for hydraulic simulation, and a React-based frontend. The AI assistant supports dashboard loading, panel operations with confirmation prompts, time range and variable management, natural language querying with SQL transparency, district-specific zooming, and undo functionality via Grafana’s version history. Development involved extensive prototyping, starting with QGIS Web Client linked to Google Looker Studio dashboards through spatial URL attributes. After platform evaluation, Grafana was selected for unified visualization. The AI assistant evolved from a CLI tool for local JSON modification into a hybrid architecture: LLMs extract user intent while Python functions perform precise actions. Key milestones include minimizing LLM use for reliability and token efficiency, integrating the Grafana API with WebSocket support for real-time interactions, React frontend implementation, and advanced functions like natural language data queries and geospatial navigation. System validation included functional testing, performance analysis, and user experience studies. Comparative results showed major improvements in user interaction efficiency and reduced task completion time, with increased accessibility for users of varying technical levels. Complex operation error rates dropped due to automated workflows and intelligent parameter checks. A real-world case study in Marene (Cuneo, Piedmont), Italy, validated the system’s effectiveness across a full-scale water network with flow, pressure, and level sensors, supported by district-specific GeoJSON files for detailed geospatial analysis. This research contributes to infrastructure monitoring, cloud architecture, conversational AI applications, and HCI design. Notable contributions include new integration models for AI-driven interfaces in industrial systems, transparent query-generation workflows, resilient cloud-native patterns, and comprehensive evaluation methods for AI-enhanced monitoring platforms. The work defines a new paradigm for making advanced analytics tools accessible to operators with varied expertise, demonstrating how conversational AI can enhance operational performance in technical environments.