• Use LCEL for both Chat and QA chains, and use multi-query retriever to increase recall
• Add running dots indicator when loading AI messages since conversational QA with LCEL and multi-query retriever is a bit slower. Show the user it's not stuck, just loading