Hopefully many of you are out there scheming on how to leverage the most recent advances in AI to create value for the world. Likely you plan on capturing some of that value. Generative models are amazing for supplementing creativity, but outside of generating text, images, and videos, one might be struck with the question about how else they could apply the amazing capabilities of these language models.

Another family of capabilities that gets less advertised, but is no less trivial to implement, is using language models as enabling the creation of fuzzy search engines. A tool that I came across on a recent quest to kick the tires of a startup idea is the open source package Weaviate. Weaviate is what is called a “vector search engine”. This is combined with most of a generative language model, such as CLIP, to allow semantic search. In addition, you can combine this semantic search easily with metadata to do semantic + symbolic search. Let’s walk through what that means with an example.

Say you’re making an app for wedding photographers. They each have thousands of pictures from many weddings stored on their computer. Lets say they were browsing through the photos, and one of a mustached relative of the bride doing the worm caught their eye, but they didn’t stash it away somewhere special. Using your app that leverages semantic search, they might be able to load in all of your photos as “vectors” into the search engine, and type in a description of the man and the dance move, and maybe narrow the search to the time of the reception using metadata.

Another possibility for vector search engines is using their automatic classification feature. Lets say