01Live02Batch03About
Target/Score · AI shot scoring

Snap the target.
Pin every shot.

Claude Vision · OpenCV · FastAPI

A hybrid Claude + OpenCV pipeline for bullseye target shooting. Computer-vision alignment, a vision model for hole detection, fractional rings and decimal scores.

100%
within ±1 ring
~4s
per live shot
<0.1s
offline mode
target · 50m precisionlive score
last shot9.2
How both pipelines work

Align · SIFT

Put the photo in template space so the known ring geometry is applicable — before anything else measures anything.

Ask · Claude Vision

“Where did a shot appear?” Claude answers with an approximate location — a hint, not a measurement.

Verify · OpenCV

Pixel-diff against the template to confirm a real perforation and measure exactly where it landed, down to the centroid.

Auto-capture · OpenCV-only · 25 fps

Continuous video, motion-triggered, scored on the edge.

Backyard rig, real shots — not a render.

Auto-capture rig · in build

The scoring pipeline it feeds ships today — try it yourself ↓

/ 01Live

Sequential scoring, one shot at a time.

Before/after pairs — Claude locates the new hole, OpenCV measures it down to the centroid.

Read the pipeline Start a session
/ 02Batch

One photo, every shot.

Claude surfaces every perforation at once, OpenCV measures them in a single shared diff.

Read the pipeline Upload a photo