Year: 2025
Type: article
Abstract: The emergence of large language models (LLMs) has opened new avenues for integrating artificial intelligence into architectural design workflows. This paper explores the feasibility of applying generative AI to solve a classic combinatorial problem: generating valid building geometries of a modular family house structure. The problem involves identifying all valid placements of six spatial blocks under strict architectural constraints. The study contrasts the conventional algorithmic solution with generative approaches using ChatGPT-3.5, ChatGPT-4o, and a hybrid expert model. While early GPT models struggled with accuracy and solution completeness, the hybrid expert-guided approach demonstrated a successful synergy between LLM-driven code generation and domain-specific corrections. The findings suggest that, while LLMs alone are insufficient for precise combinatorial tasks, hybrid systems combining classical and AI techniques hold great promise for supporting architectural problem solving including building geometry generation.
Source: Architecture
Authors: Zsolt Ercsey, Tamás Storcz
Institution: University of Pecs
Language: English
Subfield: Architecture
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,207
OpenAlex ID: W4414096393