Explainable AI for Architectural Design Generation with Few-Shot Learning

Generative design is transforming the field of architecture by enabling intelligent and adaptable solutions that adhere to both user preferences and traditional design principles. However, existing design approaches often lack interpretability and transparency, limiting their applicability in culturally sensitive domains. To address these limitations, we propose a novel framework that combines Few-Shot Learning (FSL) with Explainable AI (XAI) for generating architectural designs rooted in principles such as Vastu Shastra. Our framework utilizes prompt-engineered fine-tuning of a Stable Diffusion model on a dataset of floor plans which consists of around 50–100 floor plan designs, allowing for rapid adaptation to user-specific design requirements. To ensure interpretability and foster user trust, we incorporate SHAP (SHapley Additive exPlanations) to provide clear, interpretable insights into each design's adherence to predefined architectural guidelines. This approach not only offers personalized and culturally compliant design solutions but also bridges the gap between AI-generated outputs and user understanding, making intelligent design accessible and trustworthy.