Applying Retrieval-Augmented Generation to Construction Contract Risk Analysis

Contract risk analysis is fundamental for mitigating adverse impacts in construction projects. Yet, it remains a labor-intensive process that relies on expert reviewers. Existing natural language processing (NLP) methods can be used to automate this task, but they often fall short due to impractical assumptions, extensive preparation effort, and lack of interpretability. To fill this gap, we propose a novel retrieval-augmented generation (RAG) methodology tailored for contract risk analysis. Drawing inspiration from human review practices, this method studies historical resources with analogous clauses to produce risk intelligence. This work demonstrates its efficacy as a domain-adaptable solution, with the integration of (1) parametric knowledge from large language models (LLMs) and (2) non-parametric knowledge from domain-specific documents and legal cases. Future work will explore its applicability to other project management areas.