In complex systems such as nuclear power plants, the effect of human error cannot be overlooked, as even minor mistakes can lead to severe safety, operational, and financial consequences. IDHEAS-ECA (Integrated Human Event Analysis System for Event and Condition Assessment) is widely recognized as the most advanced third-generation human reliability analysis (HRA) method, offering a scientifically rigorous framework supported by extensive empirical data. However, its practical implementation remains highly time-consuming and labor-intensive. These challenges limit its scalability and broader adoption, particularly in time-sensitive scenarios. To address these challenges, a novel approach is proposed, integrating large language models (LLMs) with retrieval-augmented generation (RAG) techniques to enhance decision support for IDHEAS-ECA. LLMs provide the ability to process and reason through multifaceted knowledge, while RAG ensures rapid access to relevant and targeted expert domain information. Together, they enable a more robust and responsive decision-making process, significantly improving the efficiency and reliability of human reliability analysis. This approach not only reduces the time and labor required for implementing IDHEAS-ECA but also minimizes the potential for human oversight by automating key aspects of data analysis and interpretation. Moreover, the scalability of the framework suggests its potential applicability in other safety–critical industries, further demonstrating its versatility and impact. It also introduced a potential approach for dynamic real-time analysis in HRA.