| Title |
AI Agent-Based Design Approach Reflecting Expert Knowledge in Design Guidelines |
| Authors |
김하윤(Kim, Hayoon); 이진국(Lee, Jin-Kook) |
| DOI |
https://doi.org/10.14774/JKIID.2026.35.1.035 |
| Keywords |
Evidence Based Design; AI Agent; Mild Cognitive Impairment; Retrieval Augmented Generation; Generative AI |
| Abstract |
The field of spatial design is expanding toward inclusive design that accommodates users’ physical and cognitive characteristics, with spaces for Mild Cognitive Impairment(MCI) users requiring evidence-based approaches grounded in scientific research. However, traditional Evidence-Based Design(EBD) processes rely on designers’ qualitative expertise and fragmented knowledge scattered across literature and guidelines, requiring significant time and effort. This study proposes an AI agent-based system that automates the visual application of design guidelines reflecting expert knowledge for MCI users’ bathroom spaces. Unlike conventional single AI systems, the proposed system employs a multi-agent collaborative structure where specialized agents autonomously perform space analysis, evidence retrieval, design visualization, and justification provision through a four-stage workflow. The system generates comprehensive EBD reports that integrate spatial analysis results, evidence-based design recommendations, visualized improvement images, and explainable justifications with traceable sources. Validation results showed 79% semantic consistency with guidelines and 80% visual implementation of safety elements, demonstrating the system’s capacity to reflect detailed safety guidelines often missed by general generative models. This study demonstrates how AI agents can automate the translation of scientific guidelines into visual spatial representations while maintaining verifiable connections to original evidence. While currently focused on MCI users and bathroom spaces, the proposed framework presents potential for expansion to diverse user groups and spatial typologies in automating evidence-based design visualization. |