Semi-related: I guess some companies have already started augmenting their search with the use of LLMs. If you've worked on a similar project, what is your (very) high-level architecture? Did you see a noticeable difference in relevance and query time?