I've been building a quality scoring API for text content. The idea: instead of writing rules for what good looks like, each profile is trained from examples of great and not-so-great writing. The profile then scores anything you throw at it, with per-trait scores.
Two profiles live right now — compelling READMEs and technical writing. Happy to train more profiles on request. Try it:
curl -s https://qed.systems/p/qed.compelling-readme -H "Content-Type: application/json" -H "Accept: text/plain" -d '{"content":"https://github.com/astral-sh/uv"}'