Next-Track Logic for a Million-Song Radio Service Driven by One Listener Input

A radio station draws its music from a catalog of more than one million songs. Design the algorithm that decides which song plays next, given that the listener may give only one input, on the landing page. ```hint Pick the input first Decide what the single input should be before designing anything else: it sets how much the system knows when it chooses the very first song. ``` ### Clarifying Questions - Does each listener get their own stream, or is this one broadcast that everyone hears? - After the landing page, can the system still observe skips, pauses or a listener leaving, or does it get no signal at all? - Are listeners signed in with a listening history, or could each one be a first-time visitor? - What do we know about each song: artist and genre tags, audio features, play counts, licensing limits? ### What a Strong Answer Covers - The choice of the single input, weighed against the other inputs it could have been - How the catalog is narrowed to candidates and ranked for the next slot, including cold start for new listeners and new songs - Sequencing rules that keep the stream pleasant to listen to, plus any licensing or business limits - How the system learns within and across sessions, and the metrics used to judge it ### Follow-up Questions - How would you balance familiar songs against discovery, and how would you test where to set that balance? - Many listeners leave during the first two songs. How would you find out why? - A song released today has no play history. How does it reach anyone's station?