Music Discovery

Spotify's algorithm thinks you want to hear the same twenty artists forever. Apple Music's "For You" is marginally better, but it's still optimizing for engagement — safe picks, popular picks, picks that keep you streaming without ever making you stop and say who the hell is this?

Music Discovery doesn't care about engagement. It reads your Apple Music library, maps your artists against a web of musical similarity, and scores candidates based on how close they are to what you already love — not what's trending. It finds the artists that should be in your library but aren't. The ones the algorithms skip because they're not famous enough to be profitable.

First run, seven of my top fifteen recommendations were artists I already loved but hadn't added to my library. It gets better over time, too — tell it what you liked and didn't, and the adaptive engine recalibrates. Each round of feedback makes it sharper.

How It Works

Python under the hood. It pulls your library via Apple Music API, maps artist relationships using music-map.com (Playwright-driven browser automation — no API, just raw web scraping), and cross-references with Last.fm and MusicBrainz for metadata. A logistic regression model (scikit-learn) scores every candidate artist. JXA and AppleScript talk directly to Music.app for library reads and playlist creation.

The playlist builder can generate curated playlists in Music.app from your top recommendations. Fair warning: this feature works, but it's experimental. It has been known to make Music.app deeply unhappy. Evil but awesome. Don't use it without backups.