Environmental impact analysis of structural steels using retrieval-augmented generation and large language models

Reducing embodied carbon in the built environment requires reliable methods to assess structural materials, particularly steel, one of the most carbon-intensive. While environmental product declarations (EPDs) provide standardised and verified life cycle data, their integration into design workflows remains limited due to challenges in accessibility, interpretation, and cross-product comparison. Moreover, the growing volume and complexity of EPDs demand significant technical expertise for effective use with analytical tools. This study introduces EcoQuery, an artificial intelligence (AI) assistant based on a retrieval-augmented generation (RAG) framework integrated with large language models (LLMs), designed to address these barriers. EcoQuery enables natural language (NL) question answering (QA) and interpretation of EPD data for structural steel products by automating the extraction of environmental indicators and supporting reasoning tasks. EcoQuery demonstrated strong performance, achieving 83% accuracy, 87% precision, 85% recall, 81% exact match, and a Bidirectional Encoder Representations from Transformers Score (BERTScore) of 0.89, indicating high semantic fidelity. By using RAG to connect EPD data with language model reasoning, EcoQuery lowers the expertise required to work with life cycle data and supports carbon-conscious, data-driven material decisions. While demonstrated on steel EPDs, EcoQuery is adaptable across materials and sectors, offering a scalable pathway to decarbonising the built environment. • EcoQuery automates LCA data use from steel EPDs. • RAG-LLM improves accessibility of carbon data in design. • Reduces barriers in life cycle sustainability assessment. • Supports low-carbon decisions in the built environment. • Scalable AI tool for cross-sector environmental impact analysis.