We are building AI to accelerate real estate development by solving one of the hardest problems in architecture: generating buildings that are not only plausible, but constrained, compliant, and usable in the real world.

Architectural design is not a simple image-generation problem. A building is a structured composition of discrete elements: rooms, walls, openings, circulation, adjacencies, regulatory constraints, site conditions, and market requirements. Standard image models can produce visually convincing plans, but they struggle to respect these constraints. Classical optimization can encode constraints, but it does not scale well to the ambiguity and complexity of real projects.

We’re taking a different path.

Through specialized AI agents, we analyze the full scope of a project in parallel: regulations, market data, site constraints, and architectural inputs. Each dimension is retrieved, structured, and cross-referenced to build a coherent understanding of what can be built.

Gaudi-1, our first proprietary generative model, builds on this analysis using a discrete diffusion approach : a paradigm particularly well-suited to architecture, where outputs are not just pixels, but structured layouts made of discrete architectural components.

Rather than treating constraints as a post-processing step, we integrate them into the generation process itself. Gaudi-1 generates architectural drawings as structured compositions, allowing analysis and generation to inform each other from the outset.

The goal is not to replace architectural judgment, but to augment it. Our in-house architects remain in the loop, working with Gaudi-1 to reason through constraints, generate options, and accelerate the path from feasibility to design.

Gaudi-1 already reaches state-of-the-art performance on established floor-plan generation benchmarks, including RPLAN and MSD, across metrics such as IoU, FID, and KID.

But this is just the first step.

The broader challenge is far larger: architectural conception is a deeply constrained, high-dimensional problem where every decision affects the next. We are building toward AI systems that can reason, generate, and adapt inside that complexity, on real projects, with real constraints.

This is an extremely hard problem, and Gaudi-1 is an early but important step toward making architecture genuinely generative, structured, and constraint-aware.

If this sounds like the kind of problem you want to spend your time solving, join us.