This study proposes a knowledge-graph–guided retrieval method that enables large language models for structured information extraction in the nuclear domain. By integrating domain-specific knowledge graphs into LLM-based retrieval and extraction processes, the system enhances the accuracy, consistency, and organization of extracted information compared with naive LLM-only approaches. This method provides coherent, domain-aligned, and structured content that support downstream tasks such as maintenance documentation, safety assessment, and operational decision-making. This study lays a foundation for reliable AI-assisted information processing in the nuclear maintenance field.