The adoption of AI in technical fields is fueling a need for more rigorous, more modular, and more domain-specific validation frameworks. At Anori, Alphabet X’s moonshot for housing and development, we are applying multimodal generative AI to make building and development dramatically faster, less costly and more efficient. The architecture, engineering, and construction industry suffers from entrenched inefficiency due to the highly constrained, highly regulated, and regionally-specific nature of building development that currently relies on numerous domain experts to sequentially flesh out a design.

Anori is creating a unified AI platform to change the paradigm and make it possible to design for many of these constraints simultaneously. Importantly, building design requires more than just beautiful images and eloquent descriptions: it requires engineering precision, regulatory compliance, and functional performance. In this setting, it is critical to have rigorous frameworks for evaluating model outputs at each step of the process, including both automated validation and human feedback. In this talk, we’ll discuss frameworks to bring transparency, diagnosability, and specificity to the model validation process.