Reading up on SID-1

Smaller, domain-tuned LLMs trained specifically for search orchestration can replace complex monolithic retrieval pipelines by intelligently controlling simpler retrieval primitives.

Using a basic BM25 or e5 retriever with an LLM agent can achieve 0.289→0.453 NDCG on Amazon ESCI by reasoning over queries, but this approach fails on passage retrieval where the embedding model already knows best.