Journal of KIBIM ISSN:2288-1697(Print) 2288-1697(Online) 한국BIM학회논문집

Regulatory review in the early design stage is an essential yet complex task in architectural practice, as building codes often contain multi-layered delegation and cross-referencing structures that are difficult to interpret systematically. Traditional CAD- and BIM-based rule-checking systems have improved efficiency, but they remain constrained by reliance on explicitly codified rules, which fail to capture contextual linkages across different provisions. To address these limitations, this study develops an algorithm that integrates a relational database (DB) of Korean building regulations with a large language model (LLM)-based question–answering module. The relational DB structures hierarchical and referential connections among legal clauses, while the LLM module generates adaptive responses that support legal interpretation and compliance review. The algorithm was implemented within an authoring tool environment that allows textual verification of regulatory results and volumetric visualization of building mass. This dual output enabled users not only to check whether key provisions were satisfied but also to understand their spatial impact on early-stage massing design. The relational DB was constructed by focusing on six critical categories affecting building form—building coverage ratio, floor area ratio, setback distances, landscape requirements, height limits, and parking provisions. These were encoded into a structured DB reflecting the hierarchical delegation and reference patterns of Korean building codes, and then connected to the LLM module to support context-sensitive Q&A-based analysis.Validation was conducted through a survey of 29 architectural practitioners with an average of 13 years of professional experience. Respondents evaluated the algorithm for effectiveness, accuracy, applicability, and user experience. Results confirmed the potential of the system to provide simplified yet reliable analysis that can support decision-making in the early design stage. At the same time, experts highlighted prerequisites for practical adoption, including real-time statutory updates, integration of local ordinances and district-unit plans, incorporation of site-specific conditions such as land slope and adjacency, and transparent reasoning logic that reveals the legal basis of each result. This study contributes academically by demonstrating the feasibility of combining relational data modeling and LLM-driven reasoning for building code review. Practically, it provides a pathway toward intelligent, adaptive, and explainable methods that can enhance both design productivity and compliance reliability, laying a foundation for future development of scalable and context-aware automated systems in architecture.