This project introduces a performance-informed design parametrization-to-exploration end-to-end framework using neural implicit fields for building design applications. This work expands the possibilities for performance-informed building design through: demonstrating the diversity and morphological qualities of a building latent space directly conditioned on geometry with no explicit parametrization, developing a structured approach to latent space exploration for design that concurrently captures semantic characteristics of the design and performative landscape and provides control over shape generation, and finally integrating performance-driven feedback through local gradient descent of differentiable analytical equations for building shape refinement and optimization.