Large Language Models for Intelligent Enterprise Knowledge Management and Decision Support: A Comprehensive Review of Retrieval-Augmented Generation, Governance, and Deployment Architectures

Abstract

Enterprise Knowledge Management (EKM) is undergoing a paradigm shiftdriven by recent advances in Large Language Models (LLMs) and generativeartificial intelligence. Traditional enterprise knowledge systems rely heavilyon structured databases, manual metadata tagging, and legacy keyword-based search algorithms (e.g., TF-IDF, BM25). These legacy paradigms frequently fail to synthesize multi-modal unstructured documents, extract latent contextual insights, or provide accurate, context-aware decision support for executive decision-makers. This comprehensive review paper systematically examines the integration of state-of-the-art LLMs into modern enterprise knowledge infrastructure. We evaluate core technical frameworks including Retrieval-Augmented Generation (RAG), Fine-Tuning (LoRA, QLoRA), Vector Database Indexing (HNSW, IVF-PQ), and Knowledge Graph (KG) hybrid architectures. Furthermore, we address critical enterprise deployment challenges, such as hallucination mitigation, context window limitations, fine-grained Role-Based Access Control (RBAC), data privacy, and governance compliance (GDPR, EU AI Act). Finally, we present empirical benchmarking evaluations across key operational metrics, outline prominent research gaps, and chart out strategic directions for resilient, verifiable enterprise cognitive assistants. KEYWORDS: Large Language Models (LLMs), Enterprise KnowledgeManagement (EKM), Retrieval-Augmented Generation (RAG), DecisionSupport Systems (DSS), Vector Databases, Knowledge Graphs, Governance & AI Safety.