Curating an Eight-Hour Multi-Generational Holiday Playlist with AI Radio in Music Assistant

Vintage vinyl grooves blending with glowing audio soundwaves and neural AI nodes on a cozy holiday media console

     Whoever controls the speakers at a family holiday gathering has about twelve minutes before someone complains. Too loud, too old, too new, or the dreaded "haven't we heard this one already?" I wanted to build an eight-hour soundtrack that three generations could share, from grandparents who grew up on 1950s vocal jazz to teenagers who live on algorithmic streaming. Static playlists feel lifeless the second you hit shuffle, and radio stations repeat themselves before the turkey is carved, which leaves the person holding the remote as the family's unpaid, deeply unpopular disc jockey.

     My answer lives in Music Assistant, which already feeds audio around my homelab, and its AI Radio feature. Instead of treating it like a black box, I pulled open the source and read through the modules called constants, hosts, runtime, and rendering until the moving parts made sense, because I trust a tool a lot more once I know where its decisions come from. I also wrote a companion walkthrough on By Design Won't Fix, Building Your First Home Assistant App and Custom Playback Script, if you want to wire up your own playback automations.

     Here is where the fun started, because I could finally see how the engine thinks. It builds a pool of candidate tracks, scores each one against what the prompt asks for, and lets the runtime decide what plays next. Once you know there is a scoring step with a tunable variance, you stop begging the prompt to behave and start steering it like a dial on a mixing board. That shift in mindset was worth the entire evening of code reading, and it changed how I write every prompt that follows.

     My first attempts paid what I now call the systems tax. A naive prompt like "holiday party background music" drifted straight into mall-carol repetition, the same handful of jingles wearing slightly different sweaters, played with great confidence and zero imagination. When I loosened the prompt to escape that rut, the queue swung from an Ella Fitzgerald standard into a modern synth track, and the whiplash was jarring enough that I physically winced at my desk.

     So I switched to a simple two-step move: find, then take. First I found the right lane by anchoring the opening hours to 1948 through 1965, where the acoustic warmth lives: brass, vocal trios, and records that sound like they have a little dust on them. Then I took control of the pacing by tightening the candidate scoring variance and stepping the decade anchors forward into 70s and 80s soul and pop as the afternoon built energy. Those two moves tamed the repetition and the whiplash in one pass, and the queue finally felt like a thoughtful host had planned the whole day, with a natural arc from quiet morning warmth to a lively evening.

     Back to that twelve-minute rule. Instead of defending the playlist, I got to stay out of the way and let the music do the hosting. Reading the code turned a mystery box into a tool I could tune, and that is my favorite kind of upgrade because it keeps paying off long after the holidays are over. What is the one holiday music argument you would love to settle with a little engineering?

Later,
     SteveO