I don't know of any community projects using BERT atop Algolia for semantic search, but we have been experimenting with it internally. Earlier in 2020 we ran some tests that first performed a search on Algolia, before using BERT for the "semantic" side of it. We stopped those experiments for three reasons:
1. The results weren't high-quality enough 2. The predictions took too long 3. We found other approaches that performed better
We could have probably improved the quality of the predictions, and there were strategies we considered to avoid the issues around prediction time. But since we had better approaches at the ready, we put further experimentation on hold. (Apologies for being coy about the other approaches, but we are still working on those and aren't sure if they will end up in production or not.)
I'm happy to answer more questions, or if you have something you're working on that you'd like to discuss you can reach out as well. (Email is dustin@)