Model Anything in Real Time
DreamGrid
What it does
DreamGrid turns room measurements and a budget into a game-like, correctly scaled 3D space that users can decorate and shop from. Users can place, rotate, snap, stack, and inspect furniture; add windows and doors; measure clearances; and preview how the room changes throughout the day.
A furniture photo is analyzed into a constrained specification, then converted by a trusted Blender pipeline into a real-scale GLB model. DreamGrid can also find real products through Google Shopping, track the room's live subtotal, recommend cheaper alternatives, and approve the final shopping plan with a passkey before completing a Visa sandbox transaction.
Dianne's Contributions
- Scaffolded the monorepo, shared contracts, and continuous integration workflow.
- Designed the furniture specification and built the FastAPI, GPT‑5.6 Sol, Blender, and GLB-normalization pipeline.
- Built the core Three.js room experience, including placement, snapping, rotation, overlap detection, surfaces, openings, and lighting.
- Implemented spending mandates, passkey approval, signed tokens, Visa Acceptance sandbox checkout, receipts, and the payments MCP server.
The team generated correctly scaled, placeable furniture from a single product photo in about 90 seconds and demonstrated the full path from room planning through budget-aware purchasing.