Sustainability goals in construction and infrastructure projects depend on the integration and management of complex, multisource data. However, engineering management practitioners frequently encounter fragmented regulatory documentation, dispersed standards, and inaccessible sustainability guidance. This paper presents a decision-support platform, the sustainability knowledge hub, developed using retrieval-augmented generation (RAG) and large language models (LLMs). Designed to centralize and contextualize global sustainability assessment system (GSAS) requirements, the system supports project managers, consultants, and contractors by providing accurate, clause-grounded responses to sustainability-related queries. Six retrieval strategies were benchmarked across three LLMs (GPT-3.5 Turbo, Gemini 2 Flash, and Llama 3.1 8B Instruct) using the retrieval-augmented generation assessment system (RAGAS) to assess retrieval precision, answer correctness, and semantic alignment. The results reveal consistent advantages for parent-document, cross-encoder, and ensemble-based pipelines, while also quantifying latency trade-offs between lightweight and more computationally intensive configurations. A user-based pilot study with a 55-question benchmark shows that the chatbot allows practitioners to obtain correct answers more than two orders of magnitude faster than manual PDF search, with a 1.4-fold improvement in accuracy. Beyond technical performance, this paper highlights implications for engineering management practice, including improved compliance tracking, streamlined decision workflows, and enhanced stakeholder communication. Future research directions include exploring lightweight embeddings, domain-specific LLMs, and extended field deployments to broaden adoption in sustainability-driven engineering projects.
Naji et al. (Sun,) studied this question.