This project focuses on the generation of PRISM and UrbanPRISM, two large-scale simulation-aligned geometry-to-performance datasets for the built environment. It establishes a reproducible data infrastructure that systematically couples morphologically expressive building and urban block configurations with multi-modal, multi-physics performance fields. Developed through high-resolution simulation pipelines executed on HPC clusters, the datasets align structured geometric representations with environmental simulation outputs within a unified computational framework. By integrating isolated and context-aware urban settings across scales, the project enables surrogate modeling, geometry-aware performance inference, sensitivity analysis and performance-driven generative research in buildings and urban systems.