A Study on the Development of an Artificial Intelligence-Based Architectural Agent for Schematic Design Concept Using GPT-4 Language Models and Case-Based Reasoning

Abstract

This study introduces the "Design Competition Concept Proposal AI Agent," designed to reduce cognitive load and enhance creativity in schematic design. In public competitions, architects often face information overload when interpreting briefs with legal, technical, and programmatic constraints. To address this, the system integrates GPT-4-based Large Language Models (LLMs) with Case-Based Reasoning (CBR) through a retrieve–reuse–revise framework. The process involves structuring brief information, extracting keywords across five categories—sense of place, program, functionality, sustainability, and publicness—and generating three differentiated concepts guided by spatial, environmental/technical, and socio-cultural strategies. Applied to five public projects, the system produced consistent keywords and diverse concepts, with expert reviews confirming reduced processing time and improved conceptual diversity. These findings highlight the potential of LLM–CBR integration to automate conceptual reasoning and support structured human–AI collaboration in early design.