Dense vector search is excellent at capturing semantic intent, but it often struggles with queries that demand high keyword accuracy. To quantify this gap, we benchmarked a standard dense-only retriever against a hybrid RAG system that incorporates SPLADE sparse vectors.
