Real estate
housing.ai: house plans generated from your needs
A product in development at Xearth: describe your future home, get consistent plans and walk through them in three dimensions.
Xearth team1 min read
housing.ai is a product we are developing in-house: the user describes the house they want (surface, rooms, site constraints, style), and the service proposes consistent plans they can adjust, then walk through in three dimensions. The product is not launched yet; here we share what its design teaches us.
The context
Generating a house plan is not generating an image: walls must close, rooms must have usable proportions, circulation must exist, openings must respect orientation. A generative model alone produces attractive but unusable drawings. AI had to be combined with architectural rules.
What we built
- A plan-generation engine that turns needs into valid geometry, with explicit architectural rules.
- A 3D environment in the browser to walk through the result and modify it.
- A web interface to enter needs and compare variants.
- Administration and tiered billing, connected to a payment provider.
How we worked
We measured every approach before keeping it. Having a large language model draw the geometry proved far slower and less accurate than a rules engine guided by AI; so we reversed the roles: AI understands the need, the engine produces the plan, AI comments and suggests variants. An automatic quality score follows every change to the engine.
Technologies
Generation engine in TypeScript, 3D rendering in the browser, React web interface, AI services to interpret needs, Node.js API.
What we take away
AI is most useful when it frames a deterministic system rather than replacing it. It is a lesson we apply to our generative AI projects: generate, yes, but under the control of verifiable rules.