How to customize generative artificial intelligence? Case Finnish hospital construction

Purpose This research aims to propose and validate a methodology for customizing generative artificial intelligence (GenAI) to enhance its performance in a specialized domain in construction, using Finnish hospital construction as a case study. Design/methodology/approach Using a design science methodology, the study follows a three-step process: (1) problem and requirement framing, (2) solution development and (3) validation. The study applies specific research methods for GenAI customization to operationalize this framework, including domain-specific dataset curation, retrieval-augmented generation and iterative, expert-in-the-loop, qualitative validation. Findings The study indicates that a cost-effective GenAI customization can provide more relevant and useful responses than standard models in domain-specific use with approximately 120 work hours invested in customization. The validated methodology provides a transferable framework that encompasses four key stages: (1) problem and requirement framing; (2) domain-specific dataset creation, curation and sanitization; (3) model constitution and retrieval-augmented configuration; and (4) expert validation and refinement. The most resource-intensive stages were creating and curating the dataset and recruiting experts for validation. Originality/value This study offers an original contribution by demonstrating customizing GenAI for the specific localized context in construction. It underscores the importance of curated data sets and provides a validated pathway for continuous data-driven improvement.