The problem
Homeowners want a remodel quote in 30 seconds. Contractors take 3–7 days to deliver one. That gap is where 80% of remodel leads die. The estimate is the bottleneck — and the bottleneck is mostly visual reasoning a phone can already do.
The product
Open the app. Point at the room. ProEstimate detects the room type, identifies the surfaces and fixtures it can reasonably touch, and produces a render of the proposed remodel plus a confidence-bounded cost estimate. Every estimate links back to verifiable line items priced from regional contractor data.
How it works
- Capture. ARKit + LiDAR (where available) for room dimensions; a multi-view convnet for surface segmentation.
- Generation. A controlled diffusion pipeline conditioned on segmentation masks, target style, and budget tier.
- Estimation. Detected surfaces × material costs × regional labor multipliers, with hot-pluggable contractor pricing tables.
- Delivery. A shareable PDF with photo, render, line-item costs, and a CTA to book the contractor that priced it.
The hard parts
Two were genuinely hard: (1) producing renders that look like the room you actually scanned (not a stock kitchen), which required training a LoRA on 4,000 in-app captures with style targets; and (2) getting the cost variance under ±10% versus a real contractor quote, which we solved by partnering with three regional contractors who feed us live pricing in exchange for the leads.
Stack
- iOS: SwiftUI, ARKit, Vision, Core ML for on-device segmentation.
- Generation: Stable Diffusion XL + custom LoRA on Replicate; cold-start mitigated with predictive warm-up.
- Backend: FastAPI, Postgres + PostGIS for regional pricing, S3 for renders.
- Distribution: TestFlight cohorts → App Store, with RevenueCat for the subscription tiers.
Status
ProEstimate launched on the App Store in February 2026 and crossed 5,000 quotes in its first month. The contractor partner network is growing in two markets (Twin Cities, Austin); national rollout planned for Q4.