Text retrieval pipeline implementations very simple ones, going from bm25 to vector similarity search and then probably rerank.