We designed Cursor for Architecture to simplify homebuilding while reducing wasted time and materials during construction. I was one of two technical team members and led all research efforts.

We started small—working our way up from birdhouses to ADUs—which allowed us to increase complexity in both the geometry and the LLMs incrementally.

Our team explored the full potential of turning the project into a company, starting with FlexPlan as a tool for building creation. The idea was to use precomputed vendor components to enable precise material calculations—ultimately allowing for a one-click purchase option with a fairly accurate cost estimate for the home you designed. From there, we envisioned expanding into downstream supply chain management, optimizing based on real-time material demands.

The Flex LLM would learn the patterns of components, so instead of simply drawing a line in our tool, you’d be drawing a wall—with all the layers required to build it.

We used these pre-calculated components to estimate the cost and construction time of materials in real time as you designed the structure.

If you wanted to change a material, you could swap it out as easily as changing a color value in Figma—and instantly see an updated price.

This approach allowed us to translate the LLM’s understanding of patterns and building codes into a UI that was both fast and reliable—since all components were essentially pre-rendered.

Our first demo began with the most basic building blocks. We built it in WebGL and Godot to compare the value of a web-native app versus the physical realism of a game engine.

We focused on a web-native approach and started at the scale of a birdhouse, testing basic interactions inspired by The Sims.

We then integrated material calculations to begin estimating cost changes based on material choices and component sizes.

Finally, we began componentizing elements of the structure—tracking what’s inside a wall and what’s needed to meet code compliance for features like doors and windows.

In the end, it became clear that the real opportunity was to start as a data company and circle back to the design tool later. This direction didn’t align with the team’s short-term interests, so while we maintained a lightweight tool to clean some of the data we had invested in, we ultimately chose to focus on other projects.