spoti.phirios.comVibe-based playlists from your liked songs.
An NLP project that reads your Spotify library and turns a plain-language mood — “late-night drive after a long week”, “deep focus, no vocals”, “Sunday morning slow” — into a playlist that actually matches.
What it does
Pulls your liked songs from Spotify, enriches each track with audio features (BPM, energy, valence) and genre tags, then matches a free-form vibe prompt to the right subset of your library.
How it works
Track metadata via Spotify, tempo via GetSongBPM, lyrics via LRCLIB, genre tags via Last.fm. A language model embeds the user prompt and the enriched track features into the same space and ranks by similarity.
Why
Coursework for an NLP class — exploring how natural-language descriptions of mood map onto structured audio features and crowd-sourced tags.
Stack
Rust (axum) backend for the metadata service, Next.js + Tailwind for the web UI, Python for the NLP / embedding layer. Deployed on a personal Kubernetes cluster.
Status
In development. The backend track-info service is live; the bot and web UI are next. Source will be published once the homework is graded.