Structure-aware multi-agent retrieval and question answering for hydropower engineering codes

Traditional retrieval methods for hydropower engineering codes typically suffer from low efficiency, inaccurate clause localization, and limited answer reliability. To address these issues, a structure-aware multi-agent framework for hydropower engineering code retrieval and question answering is proposed. In the framework, a structured knowledge model is developed to preserve the hierarchical relationships of engineering codes and employ multiple agents to perform code routing, hierarchical retrieval, trustworthy reasoning, and answer generation; a traceable response mechanism is introduced to provide supporting code sources and clause references. Experimental results show that the proposed method achieves higher code localization accuracy, answer accuracy, and traceability than existing retrieval and retrieval-augmented generation (RAG)-based methods. The proposed framework provides an effective solution for intelligent engineering code services.