Retrieval-Augmented Intelligent Data Warehousing for Decision Support Systems

Abstract

Organizations are increasingly challenged by massive volumes of structured, semi-structured, and unstructured data generated from enterprise systems, cloud platforms, IoT devices, social media, and industrial applications. Traditional data warehouses are effective for historical analytics but lack the ability to provide semantic understanding, dynamic knowledge retrieval, and real-time decision support. This study proposes a Retrieval-Augmented Intelligent Data Warehousing (RAIDW) framework that integrates Retrieval-Augmented Generation (RAG), vector databases, knowledge graphs, semantic search, machine learning, and large language models with conventional data warehousing. The framework combines ETL pipelines, vector indexing, retrieval ranking, and transformer-based reasoning to deliver context-aware, explainable, and accurate analytical insights. Machine learning improves retrieval relevance, while feedback mechanisms continuously enhance system performance. Experimental results demonstrate significant improvements in retrieval precision, analytical accuracy, semantic consistency, query efficiency, and decision confidence compared to traditional OLAP-based systems. The proposed architecture supports diverse domains, including healthcare, finance, supply chain, manufacturing, cybersecurity, and education, while ensuring governance, scalability, transparency, and security. By grounding AI-generated responses in trusted enterprise knowledge, the framework minimizes hallucinations and enables reliable, intelligent enterprise decision support.