In our spare time, we've been experimenting with using models that we could use in our passion projects. We were testing this proof of concept of how cool it would be to have a lookup engine that relies on semantic search plus the VisionKit API. We didn't realize how many people would find this useful.

It works splendidly but without a proper mechanism to guage just how semantically similar the search query is to our stored embeddings, there could be the case where the smart lookup engine would return a result that is not semantically similar to the search query. More details on the next tab.