I work at Vectara and I'm curious -- are folks here using Retrieval-Augmented Generation (RAG)? What's your stack and what kind of improvements have you seen in answer quality?