Agentic RAG with Hybrid Retrieval and RBAC for Secure and Explainable Enterprise Knowledge Management

Enterprise organizations generate large volumes of unstructured data such as policy documents, reports, spreadsheets, presentations, and internal records. Retrieving relevant information from such heterogeneous data remains challenging due to fragmented storage, semantic ambiguity, and strict access-control requirements. Traditional keyword-based search systems often fail to capture contextual meaning, while purely semantic retrieval approaches may overlook exact keyword relevance and enterprise security constraints. To address these limitations, this paper proposes an Agentic Retrieval-Augmented Generation (RAG)-based Enterprise Knowledge Retrieval System designed to provide accurate, secure, and context-aware information retrieval across organizational data repositories.

The proposed system integrates hybrid retrieval using dense vector search (Qdrant) and sparse keyword search (Elasticsearch BM25), combined through Reciprocal Rank Fusion (RRF) and cross-encoder reranking to improve retrieval quality and contextual relevance. In addition, the framework incorporates a Plan–Act–Verify agentic reasoning loop with self-reflection and confidence-based verification to reduce hallucinations and improve answer reliability. Role-Based Access Control (RBAC) is enforced directly within the retrieval pipeline to ensure secure access to enterprise information. The system further supports ingestion of heterogeneous document formats including PDF, DOCX, XLSX, and PPTX files through a scalable processing pipeline. Experimental evaluation demonstrates improved retrieval precision, contextual relevance, and response reliability, making the proposed system suitable for enterprise-scale knowledge management applications.