Empowering SMEs with SustainWater Bot to advance urban water sustainability

Abstract

Climate change, population growth, and resource constraints are intensifying pressure on urban water systems (UWSs), prompting a shift toward integrated information management. Due to their agility and reach, small- and medium-sized enterprises (SMEs) are central to this transition. However, many SMEs lack access to robust information systems (ISs) that consolidate government initiatives, industry trends, and broader water-related data, impeding sustainable adoption. This study introduces SustainWater Bot, a chatbot driven by generative artificial Intelligence (GenAI), including large language models (LLMs) and retrieval-augmented generation (RAG). Designed to fill this information gap, SustainWater Bot addresses the shortcomings of conventional LLMs, such as information misalignment, over-complexity, and information deficiencies. RAG enables semantic consolidation of various sources, such as news, government reports, industry insights, academic research, and social media, into an integrated IS. The evaluation results showed that RAG with LLM-based methods outperformed traditional information retrieval (IR) techniques, with Llama3.2:3b achieving top scores in precision (95 %), completeness (95 %), and exact match (90 %). Traditional IR techniques such as term frequency-inverse document frequency (TF-IDF) and best matching 25 (BM25) performed lower but offered quicker responses. SustainWater Bot supports informed decision-making through a question-answering (QA) framework that delivers relevant insights on sustainable urban water initiatives (SUWIs). It is built on open-source technologies and offers SMEs a cost-effective, scalable, and sustainable solution to enhance eco-friendly water practices and operational efficiency.

Keywords Large language models (LLMs), Retrieval-augmented generation (RAG), Small and medium-sized enterprises (SMEs), Sustainable transitions, Urban water decision-making