Originally a school project for implementing shared queues in Spotify, turned into a knowledge graph visualization of music taste similarity with your friends.
You log in with Spotify, your friends log in with Spotify, and the app builds a force-directed graph where edge weights reflect how much your listening habits overlap. The closer two nodes, the more your tastes align. admittedly, some work needs to be put towards interpretability.
- pulls your top tracks and artists from the Spotify API
- computes similarity scores between you and your friends using genre/artist overlap and lyric-derived embeddings
- renders everything as an interactive D3 force graph — node opacity maps to taste similarity, you can drag nodes around, click to explore profiles
- frontend: React + TypeScript + D3, hosted on S3/CloudFront
- backend: Python Lambda functions behind API Gateway (AWS SAM)
- storage: DynamoDB for user profiles and friend graphs
the similarity mechanism is rough. the plan is to lean into listening history and recent footprints — building richer embeddings from what people actually play, not just what Spotify surfaces as their "top" content. something more continuous and less snapshot-y.
we are open to any contributions or suggestions, especially in regards to richer cooler embeddings or similarity computations (relying on waves, for instance), enhancing interactivity and interpretability of the app, etc.
Rodrigo Sastré