A Generative AI-Based Construction Safety Assistant Using Retrieval-Augmented Generation

Document Type

Conference Proceeding

Publication Date

1-1-2026

Abstract

Safety remains a central challenge in the construction industry, where OSHA standards define most regulatory requirements. Although comprehensive, these standards are dispersed across lengthy documents, making it difficult to locate provisions relevant to specific tasks or safety concerns. This complexity is especially evident in education and training settings, where instructors must interpret regulatory text and learners often lack familiarity with its structure and terminology. Recent advances in generative Artificial Intelligence (AI) offer new opportunities for improving access to technical information, but general-purpose models lack the grounding needed for authoritative regulations. This study presents a Retrieval-Augmented Generation (RAG) platform that provides clause-linked answers based on OSHA construction standards. The platform integrates document preprocessing, semantic retrieval, and citation-aware response generation through a web interface that enables direct inspection of referenced clauses. Automated evaluation using thirty representative safety questions showed strong performance, with average scores of 0.95 for context relevance, 0.99 for factual accuracy, and 0.90 for response relevance. The study contributes a verifiable and regulation-aligned approach for delivering accessible safety guidance, offering an adaptable framework for construction safety education and training.

Publication Title

Proceedings of the International Symposium on Automation and Robotics in Construction

ISBN

[9780645832235]

Link to Full Text

Share

COinS