The problem
If you own a lot and wonder "could a pool fit back there? what would it run me?" — the honest answer today is: book a site visit, wait for two or three contractors, and compare quotes that aren't scoped the same way. That's weeks of latency for what is fundamentally a geometry-and-pricing question.
Approach
The pipeline is deliberately boring:
SAT TILE ──▶ VISION MODEL ──▶ FEATURE MAP ──▶ FEASIBILITY ──▶ COST BY TRADE
(parcel) (segmentation) (structures, (pool / ADU / (excavation,
setbacks, garage, per concrete,
free area) local rules) electrical…)
- Ingest — pull the parcel's satellite tile and boundary.
- Read — a vision model segments structures, hardscape, vegetation, and open ground; free buildable area falls out of that.
- Judge — candidate improvements (pool, ADU, garage) get scored against the available footprint and setback heuristics.
- Price — each feasible improvement is broken down by trade — excavation, concrete, framing, electrical, plumbing — using regional cost baselines, so the estimate reads like a contractor's bid, not a single mystery number.
What it outputs
For each candidate improvement, a card: feasible or not, the constraining factor if not, and a trade-itemized cost range. The goal is that a homeowner can walk into a contractor conversation already knowing the shape of the answer.