Link back to content filtering notebook/recipe up top to help people who may be new or are getting started go through the easier example first?
The explanation of the range query score output as it relates to rating predictions could be clearer (should we have a helper method to convert this value?)
Consider storing the user items in a separate index to handle two things -- storage of users/metadata/vectors in a more production like way. Retriever user from DB and then do the inference? Just an idea.
Love the bloom filter approach, but I think we need to lay the baseline with just standard title/ID based filtering using a TAG filter first. And then build to why a bloom filter works here and what problem it solves.
Nice work! My feedback below, just a few things!!