A Hybrid Retrieval-Augmented Generation Framework with Role-based Access Control for Enterprise Knowledge Management

Enterprise knowledge systems operate over heterogeneous and distributed data sources, making accurate and secure retrieval challenging. This paper presents a multimodal Retrieval-Augmented Generation (RAG) system that integrates hybrid semantic retrieval, structured query execution, role-based access control, and blockchain-based audit logging. The system introduces a two-dimensional query classification mechanism for joint routing across structured and unstructured pipelines, along with a weighted reranking strategy combining semantic similarity, keyword overlap, and source reliability. Evaluation on a multi-format enterprise dataset demonstrates 87.5% top-5 retrieval accuracy and 92.1% SQL correctness, outperforming a Vanilla RAG baseline. The current evaluation focuses on retrieval and structured query performance, while end-to-end response generation is integrated at the pipeline level and will be evaluated in future work.