CARA-RAG: Civil Aviation Regulation and Authority Retrieval-Augmented Generation System

The civil aviation regulatory system is huge and complex, and the existing Question Answering (QA) system is difficult to balance the retrieval efficiency and generation quality. This paper introduces CARA-RAG, which improves the reliability and authoritativeness of regulation QA by combining customized corpus construction with Retrieval-Augmented Generation (RAG), and reinforced by security verification mechanism. A domain-specific evaluation dataset was built, combining automatic generation and expert verification to ensure data quality. Experimental results show that, CARA-RAG has better relevance, accuracy, Completeness and Faithfulness than the LLM-only generation and naïve RAG models. The system also integrates a lightweight WebUI to realize real-time interaction and visual display, providing a complete and scalable solution for civil aviation regulation QA system.