From hand-drawn floor plan to furnished 3D model and AI layout guidance.
The challenge
Most AI room tools ask for a photo and return generic advice. This product needed a workflow that let users describe a room precisely, understand why the guidance mattered, and compare alternatives before rearranging anything in the real world.
How we built it
I designed a 9-step flow — draw, refine, transform, annotate, furnish, preview, analyze, results, and optimize — wrapped in a vintage sumi-e ink-wash design system. Skia handles the freehand drawing surface, Three.js renders the space, and a Firebase Cloud Function calls Gemini server-side (the API key never ships to the client), returning structured evaluations across bagua zones, element balance, chi flow, command position, and a harmony score with ranked recommendations.
What shipped
The result is a spatial AI product with real depth: users annotate walls and fixtures, place from a 72-item catalog, preview recommendations in motion, and choose between scored layout alternatives instead of guessing at a text-only suggestion — with a Room Studio beta adding inline 2D/3D editing and undo/redo.
Outcomes
- Built a 9-step pipeline from freehand Skia drawing to simplified geometry, room annotation, furnishing, and a furnished 3D preview.
- Created a 72-item furniture catalog across 11 categories plus a wall, floor, and ceiling annotation system for doors, windows, outlets, vents, and lighting.
- Added server-side Gemini analysis (an overall score plus chi-flow, element-balance, command-position, and harmony sub-scores) with animated layout optimization that offers 2–3 scored alternatives before the user applies changes.