GIS-MX01
Their edge is being early on a parcel. Getting to an opinion on one meant an hour across nine public websites.
- Client
- Land acquisition, SeattleWithheld by agreement
- Year
- 2026
- Role
- Product, design & engineering
- Type
- Internal tools
- Approach
- GIS pipelines · Retrieval · Web app
Where it started
Acquisitions ran out of a browser with a dozen tabs open: the county parcel viewer, the city zoning map, permit records, assessor history, listing feeds, and a spreadsheet holding it together. Working up a single parcel took most of an hour, and half of that was retyping numbers from one site into another. Whatever they didn’t get to that week, somebody else got to first.
What was hard
The data is public. It is not comparable. Sources disagree on boundaries and lot area, layers refresh on their own schedules, and the questions that decide a deal — what can go here, how much of it, how tall — are not a field in any dataset. They are prose in a municipal code that gets amended. A model asked to recall zoning answers confidently and sometimes wrongly, and a confident wrong answer here means bidding on a lot you cannot build on. Everything the system says has to be checkable against a source in one click.
What we built
Nightly pulls from Seattle and county open data and GIS layers, normalized into one parcel record with the disagreements between sources shown rather than averaged. A watch that scans new listings, permits, and status changes against their buy box and surfaces matches the morning they appear. Land use questions answered by retrieval over the actual code text, with the section quoted on screen. Massing and mock site plans generated from lot geometry, setbacks, height limit, and FAR, so a parcel can be argued about before anyone pays for drawings.
What changed
A parcel is a page instead of an hour. Candidates arrive before the team goes looking for them. Massing is in the first conversation rather than after a consultant. Analysts still read the code section, because it is sitting right there, but they are checking one answer instead of assembling it from nine tabs.
Scope
- One parcel record from city, county, and GIS sources
- Nightly refresh with change detection
- Watchlist scanning new listings and permits against their buy box
- Land use answers with the code section cited
- Generated massing and mock site plans from lot geometry
- Exports that drop into the deal memo