AI Music Studio
An autonomous music pipeline: trend discovery → LLM brief → generation → cover art → mastering → human approval.
Problem
Running virtual artists means repeating the same chain for every release: research trends, write a brief, generate, pick takes, make artwork, master, tag and publish.
How it flows
- Discover
- Analyze
- Brief
- Generate
- Audition
- Artwork
- Master
- Approve
- Publish
What I built
- A trend radar built from live chart sources, matched to artist personas (language, sound, visual identity).
- Per-task model routing for briefs and scoring; generation, auditioning and download of takes.
- AI cover art, loudness mastering, tagged MP3s and release packages.
- A dashboard with run timelines, live logs and playback — and a mandatory human approval gate before anything is published.
- A plugin system for new sources and services; credentials encrypted at rest.
Stack
TypeScriptpnpm monorepoGeminiOpenRouterSunoFFmpegPlugins