How to reduce OpenSearch RAG costs: query planning, batching and caching
The fastest way to reduce OpenSearch costs in a RAG system is to stop paying for retrieval you did not need: measure how many OpenSearch requests one user message triggers (queries per turn), plan and batch those requests instead of firing them as the pipeline thinks, rerank once on a
