Progressive web app · 2026
Boulder Beta
A Moonboard for any wall. Photograph a spray wall, let the app find the holds, and get a graded route drawn on top of your own photo.
What it is
Take a photo of a climbing wall. The app proposes where the holds are, you correct it with a few taps, then you pick an angle and a difficulty and it generates a route with exactly the holds you are allowed to use highlighted on the image.
Why I built it
A Moonboard is a standardised wall with a shared library of thousands of problems, which is why it is fun forever. Every other wall in the world is a pile of holds with no problems attached. The interesting question is whether you can bring the Moonboard experience to an arbitrary wall using a phone camera.
How it works
Capture, downscale on device, then a hybrid detection step. Claude vision proposes candidate holds with position, type, and size, and you fix what it got wrong, because a fully automatic detector that is right 85% of the time is more annoying than a semi automatic one you can correct in five seconds. Detection sits behind a single API contract so a dedicated segmentation model can replace the vision call later without touching the interface.
Where it stands
Thin slice MVP, the full loop from photo to highlighted route works. Grading quality is the open question and it needs real climbers to tell me whether a generated V4 feels like a V4.