Music app
2026
Livenotch.fm
Rate your music once, then ask it real questions.
Private repository
Overview
A music library built around your own opinions. Score a song out of ten while it plays, tag it by mood, vibe and sound, then ask for what you want in plain English and get a queue of songs you already love. Works with Apple Music today, with Spotify on the way, or with no music service at all.
My role
Creator. Product, design, web app, API, data and infrastructure.
Built with
- Next.js
- React 19
- Tailwind CSS 4
- Motion
- Express 5
- Postgres 16
- Drizzle
- Anthropic API
- Stripe
- WebAuthn
- Azure Container Apps
In numbers
- commits in three months
- ~1,040
- custom static checks
- 160+
- a score per song, in one move
- 1–10
How it's built.
Ratings in one keystroke
Number keys score the track that is playing. A new library starts from guesses seeded by listening history and community data, clearly labelled as guesses until you confirm them.
Ask in plain English
Smart search on the Anthropic API with structured outputs: “play upbeat country rock” builds a queue and starts it. Five kinds of request are told apart, each with its own eval scripts.
One service, or none
A provider layer for Apple Music, Spotify or no service at all. What a provider's API really allows is measured at runtime, and likes and playlists are written locally first, then synced.
Security without shortcuts
Streaming-service tokens are encrypted on the server with AES-256-GCM and never reach the browser. Passkeys guard the admin panel and a bot check guards sign-up.
Social, and measured
Comments pinned to a moment in a song, a live listening feed, 50 levels and 25 badges, and 160+ custom static checks for contrast, touch targets, layout overflow and pricing-copy drift.
Screens
notch.fm's one-move rating slider, with mood, vibe, sound and genre tags notch.fm rating the song that is playing, beside a plain-English request for a queue
The notch.fm home page on a phone