Inspiration
Designed primarily for architecture students but also architects and any other person who wants to design a building, SketchMetrics focuses on education over automation. By flagging critical mistakes and optimization opportunities—and explaining the design principles behind them—it builds users’ critical-thinking skills and helps prevent over-reliance on AI-generated outputs.
What it does
SketchMetrics turns floor-plan images into a structured building model. Users import plans, assign floors, straighten photographed drawings, set scale by selecting a known edge, review rooms and openings, and explore insights such as circulation, visibility, and sun exposure. Projects are saved locally, and users can start a new project without losing previous work.
How we built it
We built the desktop application with Kivy for the user interface, FastAPI for the local application service, OpenCV for image processing, and TopologicPy for geometry and spatial analysis. All core technologies are open source.
I created the initial user journey and development plan, then used Codex with GPT-5.6 Terra in extremely high reasoning mode to turn it into modular, gated implementation steps. The AI helped trace each change across the interface, backend, project data, tests, and documentation.
Challenges we ran into
The hardest part was vectorizing a photographed floor plan, rather than the architectural domain specifics. Photos can be skewed, unevenly lit, incomplete, or taken from an angle. Converting those pixels into trustworthy walls, rooms, and openings requires image correction, calibration, geometry checks, and clear user review points.
What we learned
A guided workflow matters as much as the analysis itself. Removing the starting grid and letting users set scale from a selected edge made the experience much simpler. We also learned that keeping automated tests and a living implementation plan is essential when a feature affects multiple parts of an application.
What's next for SketchMetrics
Next, we want to make setup even easier with a one-click Windows desktop build, improve automatic photo vectorization, and refine the review tools for walls, rooms, and openings. We also see potential for richer reporting and more advanced building-performance insights.
Log in or sign up for Devpost to join the conversation.