TileGPT: Generative AI for Intuitive Design Exploration and Trade-Offs Navigation Title

Designing for net-zero in Architecture Engineering and Construction (AEC) is complex and challenging. It requires satisfying multiple stakeholders, balancing conflicting goals and relying on slow and difficult-to-use lifecycle assessment tools. With this paper we introduce TileGPT, an innovative human-AI interaction system for direct and iterative design synthesis, trade-off navigation between conflicting metrics and expanded user accessibility, critical features to make sustainable design more attainable and scalable. The system integrates outcome-driven design optimization, procedural geometry generation with constraint enforcement and Large Language Models (LLM) to correlate relationships between high-level performance and low-level design features. The methodology provides an interactive interface for granular in-the-loop design editing, all while meeting stringent architectural and engineering constraints and requirements. The work is presented through a real-world case study for low-carbon, multifamily housing development with a US based modular building manufacturer.