AI-powered music discovery PoC on the Spotify Web API — LLM as the recommender
| .github/workflows | ||
| backend | ||
| eval | ||
| frontend | ||
| .env.example | ||
| .gitignore | ||
| AGENTS.md | ||
| CLAUDE.md | ||
| docker-compose.yml | ||
| README.md | ||
discovery-by-llm
Music discovery with an LLM as the recommendation engine, on the Spotify Web API.
Thesis: Spotify removed its recommendation and audio-intelligence layer from the API (recommendations and audio features are gone for new apps). This proof of concept asks whether an LLM can take over that role, with Spotify demoted to resolver, verifier, personalisation source, and playback surface. A naive-search baseline makes the answer measurable.
Work in progress — this file is filled in during the build. Chronological build log (Dutch): docs/logboek.md.
Demo
(to follow: hosted instance + local docker compose up --build, with and
without API keys)
How it works
(to follow: pipeline diagram and module map)
Choices
(to follow)
Performance and optimisations
(to follow)
Where to look
(to follow)
Method
(to follow)
What I cut / what I would do next
(to follow)