Why might you use a "Cross-Encoder" re-ranker after your initial vector retrieval? What is the trade-off in terms of latency?

How would you architect a retrieval system to solve this 'Multi-hop' problem? Example question: How does the Q3 revenue of our Tokyo office compare to the bonus of the CEO?

Retrieval adds a "hop" before generation. How would you minimize the time-to-first-token (TTFT) for a user?

Why is "fixed-length" chunking often insufficient? How would you handle a document where a single sentence contains a critical fact but spans a chunk boundary?