Generative AI · Vertical B2B
Remodel Studio
Photograph the room. See the after. Price it.
A contractor photographs a room; the app renders the remodel, prices it from a real cost book, and schedules the build.
- Role
- Solo — product & engineering
- Year
- 2026
- Built with
- Expo / React Native · Img2img (adapter) · Supabase Edge Functions · Pure-TS costing engine
How it works
Capture
Photograph the room exactly as it is today.
Render
Img2img keeps the room and changes only the finishes.
Price
A line-item estimate and a sequenced build schedule.
The problem
A general contractor wins work at the kitchen table, but the gap between ’imagine new finishes here’ and a credible, priced proposal is hours of back-office work — and the client can’t see what they’re buying.
The build
Walk into a house, photograph a room, set the scope. The app renders the ’after’ from the actual photo, prices it as a line-item estimate from a regional cost book with the contractor’s margin made explicit, and sequences a build schedule with trade dependencies and a critical path. One tap sends the client a proposal they can accept online.
The AI technique
The render is structure-preserving image-to-image: a hard preservation clause keeps the same room, walls, and camera angle and changes only the selected finishes — the same room remodeled, not a different room. Room dimensions come from photo-AI measurement anchored on standardized elements (36-inch counters, 80-inch doors), defensively parsed and clamped. The defensible core is a pure, side-effect-free costing and scheduling engine; AI lives only at the boundaries, and the image-model key never leaves the server.
The outcome
A verified estimating engine (every quote asserts its own invariants), a draggable before/after render, and a client-facing acceptance flow — a generative experience wrapped around deterministic money math you can trust.
A closer look
