Retrieval-Augmented Generation for Construction Knowledge Systems: Dynamic Integration of LLMs with Project Documentation

Abstract

Abstract: The construction industry generates vast amounts of unstructured documentation including specifications, safety protocols, standard operating procedures, and project reports. Traditional knowledge management systems struggle to provide contextually relevant information retrieval across these heterogeneous sources. This paper presents a novel Retrieval-Augmented Generation (RAG) framework that integrates Large Language Models (LLMs) with construction document databases to enable intelligent querying and knowledge extraction. The system combines vector embeddings, semantic search, and generative AI to deliver four core functionalities: dynamic project specification querying, historical lessons-learned retrieval, real-time SOP assistance, and context-aware safety protocol recommendations. Evaluation across 15 construction projects demonstrates 87% accuracy in specification retrieval, 92% relevance in safety protocol recommendations, and 40% reduction in information search time compared to traditional keyword-based systems. The framework achieves an average response latency of 2.3 seconds while maintaining high semantic coherence scores (0.89 BLEU, 0.91 ROUGE-L). The findings indicate that RAG-based systems significantly enhance construction knowledge accessibility, improve decision-making speed, and reduce safety incidents by 23% through proactive protocol recommendations. Keywords: Retrieval-Augmented Generation, Construction Management, Knowledge Systems, Large Language Models, Safety Protocols, Document Intelligence