WaterRAG: A Multiagent Retrieval-Augmented Generation Framework to Support Water Industry Transitions to Net-Zero

Achieving net-zero carbon emissions in wastewater treatment requires a complex, interdisciplinary information integration. While large language models (LLMs) have the potential to break down information silos, their capabilities remain limited in specialized wastewater domains. Therefore, we introduce WaterRAG, a multiagent retrieval-augmented generation (RAG) framework that couples LLM reasoning with verifiable wastewater knowledge. WaterRAG integrates a selective database of 7637 peer-reviewed studies and 11 engineering references on wastewater treatment. Through iterative collaboration among retrieval, review, and evaluation agents, WaterRAG produces evidence-based output for three critical applications in wastewater management: (i) wastewater expert-level question answering, (ii) literature review on specific topics, and (iii) plant-specific engineering scientific support. In benchmarking of 370 technical questions, WaterRAG achieved an 80.5% answer correctness rate on professional wastewater treatment questions, outperforming standalone GPT-4.1 (64.9%). WaterRAG also generated a more comprehensive citation-supported review, with quality improving across refinement iterations. Ablation experiments confirm that the superior performance arises from the synergistic contributions of optimized retrieval and iterative multiagent framework. This work represents an early domain-specific application of multiagent RAG to wastewater treatment, highlighting the potential of retrieval-grounded LLM systems to complement professional expertise and support evidence-based decision-making toward sustainable wastewater management.