Vibe Splitter

Personal side project: built and actively maintained outside of university. A self-initiated full-stack Python web app for automated Spotify library management.

Web app that automatically clusters a Spotify library into vibe-based playlists using Last.fm genre tags, MiniLM semantic embeddings, and optional Spotify audio features. Runs hourly, routes new tracks via cosine similarity, and features a real-time SSE-powered dashboard.

Key Highlights

01.Semantic clustering pipeline using fastembed (ONNX MiniLM) to encode Last.fm tags and track metadata into vector embeddings, with optional Spotify audio feature fusion.
02.Confidence-based routing: tracks above the auto-assign threshold go directly to playlists; uncertain tracks land in a review inbox for manual approval.
03.Drift detection monitors cluster cohesion over time and flags automatically when a full recluster is needed.
04.Server-Sent Events (SSE) bus powers a real-time dashboard: sync status, track counts, inbox queue, and override controls update live without polling.
05.Circuit breaker pattern in the Spotify API client handles rate limits and transient failures gracefully.
06.Containerised with Docker and deployed on Render; SQLite state layer with atomic writes ensures safe concurrent access.