Dream

Dream is a research-and-product lab based in San Francisco that trains generative models over structured, physically grounded representations of the built world. Its work sits at the intersection of generative AI, spatial reasoning and physical modelling, applying frontier machine-learning techniques to the architecture, engineering and construction (AEC) industry - a multi-trillion-dollar market.

The company's principal product is an architectural and engineering intelligence system: a set of generative models that encode the built environment as structured, relational, physically grounded data. Rather than producing visual renderings, the system reasons over constraints, relationships and design intent to generate 3D layouts that are structurally coherent and, in the company's formulation, actually buildable. Technical domains in scope include 3D layout generation, structured relational data representation, spatial reasoning and structural modelling.

Dream operates as a focused organisation rather than a sprawling one, combining frontier ML research with product experience. The team is interdisciplinary, spanning engineers, researchers, product leaders and architects, with an emphasis on dimensional precision and real-world buildability over aesthetics. The stated mission is to decode and model the built world.