curious if the above would scale to multi dimensional indexes, like they hint at the end. For things like similarity search a "learned" K-D tree might be worthwhile