34
NYC Soundhood
Connect Spotify, get your top 3 NYC neighborhoods — by ear.
"What kind of music are you into?" is a question people answer with a vibe, not a vector. Spotify already knows the real answer — it just hands it back as a Wrapped slideshow once a year. The idea here was to turn that latent taste data into something playful and place-based: if your listening were a New York neighborhood, which one would it be? The constraint was the same one every cloonk tool holds to — it had to run with no backend and never hold onto anyone's data.
A single-page app on Spotify's Authorization Code + PKCE flow — fully client-side, no secret, no server to send a token to. It pulls the user's top 50 artists, classifies each artist's Spotify genres into ten sonic families via keyword rules, weights by listening rank, and reads average artist popularity as a mainstream-to-underground axis. The resulting taste vector is matched to fourteen hand-profiled neighborhoods by weighted cosine similarity, blended with axis proximity. Results show match percentages, the driving genres, and a full taste-fingerprint readout.
- Connect Spotify → your top 3 NYC neighborhoods with match scores and reasoning
- A taste fingerprint — ten sonic families, a mainstream-to-underground dial, and your top genres — rendered from real listening
- Auth and analysis run entirely in-tab; the token is never persisted and no listening data leaves the browser
- A sample-data preview so the engine is explorable without an account
PKCE means a 'connect your account' toy needs no backend
The reflex with anything that touches a user's Spotify is to stand up a server to hold the client secret. The Authorization Code + PKCE flow removes that entirely — the browser proves it started the request with a hashed verifier, no secret required. So the honest privacy story and the architecture are the same one: there's no server to send your listening history to, because there's no server at all.