Towards Bridging Shape and Graph Grammars Through Topology: Fundamental Shape Processes for Architectural Representations in Artificial Intelligence

The emergence of transformer models has revitalized AI research in architectural design despite 3D generative AI lagging behind its 2D counterparts due to dataset limitations. Lightweight frameworks like Topologic enable rapid creation of 3D datasets suitable for graph machine learning, leveraging the library's ability to produce vector-based geometric, topologic, and semantic models. Our paper presents a topology-driven graph grammar generating architecturally meaningful 3D models that contribute to generative AI training datasets. Bridging 3d shapes, topology and graphs, our system applies shape transformation rules procedurally to create 3D models from simple graphs. This approach bridges shape grammars with topology, maintaining semantics while generating novel solutions suitable for 3D generative AI.