MongoDB's implementation separates the vector index runtime from the transactional processing enabling independent scaling and workload isolation but preserving unified query richness and scale-out via sharding. This is a best of both worlds in my view...