LLM rerankers for production RAG: tips and tricks (fin.ai)

Surprised that removing spaces actually had such a big effect on latency.

Also props for including the prompt and AB results