How we use AI

Last updated 6 October 2026

AI is part of how PeachCut works and of how this site was made. Here is exactly where.

In the product

• Reading the scoreboard. PeachCut samples frames from your game video and sends the scoreboard area to an AI vision model (Anthropic's Claude) that reads the score and the clock. When it fails on a game, Google's Gemini is tried as a back-up. We are training our own small model for the same job, so over time fewer frames leave our servers.
• Finding layouts. On a scoreboard we have never seen, an AI model locates the score and clock boxes once.
• What AI does not do. It does not recognise faces or identify players, does not decide anything about you and does not generate or alter your footage. Clips are cut from your original video; you can check and change every clip in the editor.
• Training. Scoreboard crops (digits only, no people) are used to train our model. A person confirms each training label; AI suggestions are never accepted automatically. Anthropic does not train its models on data we send through its API.

Because detection is automatic, results can be wrong. Check a reel before you publish it. The privacy policy lists every company that processes your data.

On this website

• The game footage on peachcut.com is real: games from Bulgarian leagues, cut by PeachCut itself.
• The short animated portraits in the "Who it's for" section were generated with AI from still images. They show illustrative characters, not real coaches, scouts, players or parents.
• The site's text and code were written with the help of an AI assistant (Claude). Every line was reviewed and is the responsibility of Georgi Diyanov Tsonev.