McCracken County · Kentucky · pilot
Pick any location in the county. Parcel ID, street address, dropped pin, drawn polygon. We'll surface a deep dossier on it — legal stack, terrain, drainage, neighbors, every layer queryable — in less than a day.
That's the bar: full geometric depth on any spot in our service area, same-day turnaround. The two demos below were each delivered same-day, parcel resolution to live page — the bake itself is fast enough that the slow step is the writing, not the computing. McCracken is the pilot — 35,055 parcels we can light up the same way for any customer, on the same workflow whether the lot is a 0.14-acre downtown parcel or a 4,300-acre commercial corridor.
Same workflow, two scales. One is an entire commercial corridor, the other a single downtown lot:
Toggle layers in either map: 3-inch ortho · OSM streets · dark base · parcel boundaries · roads · neighboring buildings · ridge-valley curvature · "tap to make it rain" drainage simulation. Both pages went from "let's do this AOI" to live URL in a couple of hours.
Outside the county, we've taken the same workflow over a different terrain class to prove the bake holds at scale: /turkey-bay — the OHV side of Land Between the Lakes (Trigg County), with gravity-mode terrain and a multi-mile trail strip.
The demos above show the raw data model at specific AOIs. These are the iOS apps that make the same 35,055-parcel database usable on a phone — tap-first interfaces to the legal layer, terrain, flood risk, and crop history.
Pick an AOI — by parcel ID, address, lat/lng radius, or a drawn polygon. Our service pulls the legal layer for the parcels in scope, then runs five layer bakes in sequence: vector overlay export, ortho clip, drainage flow field, ridge-valley map tiles, and a unified-map composition. The slow step is the drainage integration over a 60-cm grid; everything else is fast enough not to matter. From "give me a dossier on X" to a live page typically lands the same day.
The output is a self-contained asset bundle that drops into a templated page — every layer toggleable, every layer over the same base map, everything pannable and zoomable down to the inch.
We don't stop at "the parcel exists and has these bounds." We interrogate it in every dimension we have:
Polygon-level topology. A parcel's neighbors aren't just "what's listed in the tax roll" — they're everything that shares a line with this geometry, including unassessed rights-of-way. Public alleys don't have parcel IDs, so they don't appear in adjacency queries unless you look for them: small notches at lot corners, straight gaps that align across a block, narrow strips of un-deeded space between rows of lots. We caught one on the Broadway-315 demo — a 7-foot notch at the NW corner of the lot, marking where a Jefferson Street alley terminates against the rear boundary. Same query class surfaces shared driveways, encroachment, and platted-but-not-built easements.
The LiDAR returns plus a baked tensor field of the surface. Bare-earth and surface elevation at 60-cm resolution, with a 5-band raster of the first and second derivatives layered on top: slope-x, slope-y, curvature-x, curvature-y, and the off-axis twist component.
Most "terrain" tooling stops at slope and aspect. The tensor field carries everything past that — peak, pit, ridge, valley, saddle — without re-deriving per query. Ask "where do ridges run? where are the saddles? which parcels straddle a watershed boundary? where does the surface change from convex to concave?" and you're reading off a single field, not running a new analysis. Parker's monkey-saddle (where three of the four parcels meet at a critical point with Hessian determinant ≤ 0) was found by clicking on the tensor layer and asking the field directly.
The drainage simulation stamps canopy and structure heights as obstacles before integrating, so simulated drops respect what's actually on the ground rather than ghosting through tree canopy or buildings. That's the "tap to make it rain" interactive on both demos.
Multi-year imagery (state-flown 2024 + earlier vintages), multi-year land cover, multi-year ownership, multi-year assessments. Change detection is mostly latent: nothing is automated yet, but the data is aligned and queryable per pixel and per parcel across vintages. "Did this parcel grow a structure between 2018 and 2024? Did the canopy here come down? Did ownership change in the gap year of an assessment?" are all single queries against the same model.
The point: "tensors or GTFO" isn't a vibe — it's the architectural commitment that produces sentences like "the alley off Jefferson is traceable as a 7-foot notch in 315's NW corner" or "Parker's three-parcel critical point is a true monkey saddle of index −2." Ordinary stacks can't answer those questions because they don't carry the second derivatives, the polygon detail, or the cross-vintage alignment.
Honest list:
A property intelligence stack should answer questions, not hand back datasets. "What does my watershed look like, and which neighbors am I coupled to hydrologically?" is one question. So is "if a tax bill goes unpaid, what is the on-the-ground exposure — buildings, hardscape, canopy, drainage routes?" Both live in the same model. Ask us either one for any spot in the county and we'll have you a dossier the same day.
McCracken is the pilot. Sister counties — Ballard, Graves, Livingston, Marshall, Trigg, Lyon — come online next as we extend the same service westward.
Want a dossier on a specific spot, or this lit up for your service area? Fork Development Corp + Smol Nerd Farm — info@jays.bio.
— end —