That's a fairly clever idea, but far from novel. Face detection and image classification has been around for a while (I've used digikam's for half a decade now). For text, I will admit that semantic search is pretty helpful as well, but you can just use semantra which runs entirely on device. Personally, I run bge-micro locally on my notes. It's a marginal improvement to just using ripgrep. I expect semantic search to become the standard for file search sometime in the next decade, but right now the vector embeddings take too much space and are too costly to generate for most usecases.