This sounds very cool. I recently thought of another approach for song recommendation due to necessity i.e.: based on how a song sounds. There are genres but they're way too generic. Other methods like similar artists, what else people who've likes a song love etc doesn't always work. Not sure how streaming services recommend but I don't think they do what I want.

I know, how something _sounds_ is subjective but I was thinking something along the lines of how Shazam app does. If they can compute a song signature and compare it against a library, it should be possible(?) to find top 10 closest neighbors.

BTW, I'd love to test your app if that's possible.