Inspiration
Finding the best spot to shoot from is slow and expensive. To be sure you have the best one, you'd have to check every possible spot. Scouts do a version of that by driving around, but the real fix is to narrow down which spots are worth checking, cheaply and early, until only the best ones are left. That helps a filmmaker chasing a sunset over the skyline, a photographer, or a tourist looking for a bench with a view.
What it does
You tell our agent what you want in plain English, which it turns that into goals you can see and edit such as: sliders, must-haves, and a map layer for each goal. It searches thousands of points in Los Angeles or Princeton and gives back a short ranked list. For each spot you get the sun's path over the skyline in 3D, a Vexcel aerial scan of the surroundings, and, from CoStar, contact info for the people in the buildings.
How we built it
Starting from Vexcel's elevation data, a 1-meter height grid of every roof, tree, and street, a small Rust engine works out cheap facts about each point on a grid in the area. It looks at sun visibility, openness to the sky, elevation, distance to objects detected in your prompt; our agent reads your prompt and writes goals as a cost function over those facts and the engine runs those functions through a funnel. The "must" checks go first, the useless info is discarded.
These cost functions are considered, tested, and saved if proved to be reliable. As cost functions accumulate in our library, future searches become less and less expensive, as more and more locations can be proved deterministically.
Challenges we ran into
We had to replicate the sun positions exactly, and we used NOAA's exact formulae
Vexcel first-loading took immense amounts of time (~160s), which we eventually brought down to ~0.1ms with caching across subsequent page loads
Vexcel initially didn't have support for the area around Princeton and we scrambled to get a backup mapper working but Vexcel graciously stepped in just in time to make all that work redundant because they just geofenced the Princeton area
Accomplishments that we're proud of
Successful integration of Vexcel API with the front-end, which proved to be a significant challenge.
What we learned
What's next for golden
Weather as a metric Higher quality maps
Built With
- axum
- costar
- rust
- tokio
- vexcel