The PhD research frames this challenge as a curriculum: defining what a machine should learn, what examples it should learn from, and how that learning can be guided and evaluated. Through this approach, the work investigates how architectural knowledge can be structured and shared in ways that allow machines to recognize, relate, and frame spatial ideas. 

Situated within the broader context of multimodal AI, the project examines the tensions between visual similarity, spatial logic, conceptual meaning, and linguistic description. It argues that these tensions are not merely technical limitations, but reflect deeper differences between how architectural knowledge is produced and how computational systems represent information. Addressing this gap requires structured datasets and annotation strategies that are grounded in architectural expertise and interpretive practices. The project develops a multi-layered ontology for structuring and annotating architectural datasets, that reflects architectural ways of knowing and visual–spatial intelligence. 

The research methodology combines four interconnected components. An interdisciplinary literature review brings together perspectives from spatial cognition, design studies, and computer vision to establish conceptual foundations. An archival study analyses a dataset of approximately 100 single-family houses, using visual, textual, and metadata sources to identify recurring patterns in architectural representation. A reflexive component incorporates the researcher’s own architectural training and positionality, producing annotated drawings and interpretive visualisations that make tacit knowledge explicit. Finally, semi-structured interviews with architects examine how domain-experts interpret drawings, generating comparative insights into expert reasoning.

By integrating these components, the research develops a structured framework that bridges human interpretive knowledge and computational representation without reducing one to the other. The outcome is an annotation approach and ontology that could enhance the usability of digital architectural collections, supporting architects in analysis, precedent study, and design reflection. In this way, the project contributes to a more critical and grounded integration of computational tools in architecture, where AI supports, rather than replaces, human expertise and interpretive judgement.

This text outlines the general direction of the PhD research. The project continues to evolve alongside related topics presented on this website. The title remains provisional and may be updated as the research develops further.