They have absolutely the best take on embedding search + normal faceted search. Here is a blog post about how it works:

https://blog.vespa.ai/vespa-hybrid-billion-scale-vector-sear...