| Title |
LLM-Based Design Automation for Defining Buildable Envelopes - Simulating Daylight Setback Regulation Changes |
| Authors |
변상훈(Byun, Sanghoon) ; 오정석(Oh, Jungseuk) ; 강범준(Kang, Bumjoon) |
| DOI |
https://doi.org/10.38195/judik.2026.06.27.3.5 |
| Keywords |
설명가능한 인공지능(XAI); 설계자동화; 대규모 언어 모델(LLM); RASE 방법론; 일조사선 Explainable AI(XAI); Design Automation; Large Language Model; RASE Methodology; Daylight Setback Regulation |
| Abstract |
Automated architectural and urban design, especially when based on generative AI, is constrained by the lack of transparency in how design outputs are generated. This is problematic in practice, where design conditions must often be reconfigured in response to changing regulations. This study proposes a three-stage automated design pipeline in which the reasoning process can be verified and revised. Regulatory texts are collected through Retrieval-Augmented Generation (RAG), structured in JSON format using the RASE (Requirement, Applicability, Selection, Exception) methodology, and translated into geometric computation scripts. The method was validated through parcel-level 3D code interpretation and buildable-envelope generation for 21,584 parcels in Gangnam-gu, Seoul, focusing on the revision of the daylight setback regulation. The results show up to an 8.1% increase in buildable area in the four- to five-story range. The proposed pipeline demonstrates the potential of a transparent and interpretable tool for regulatory adaptation and automated design exploration. |