Data & AI

Hybrid Search: Dense + Sparse

Data & AI Pipelines diagram · made in Skemio

Hybrid search diagram showing user query split into dense embeddings and sparse BM25 tokens, retrieving from vector and keyword indices, then fused via RRF into final results.
Hybrid search diagram showing user query split into dense embeddings and sparse BM25 tokens, retrieving from vector and keyword indices, then fused via RRF into final results.
Open it in SkemioMake your own

What it shows

  • Dense retrieval captures semantic intent via high-dimensional vectors.
  • Sparse retrieval (BM25) guarantees exact keyword, ID, and term match.
  • Reciprocal Rank Fusion (RRF) merges rankings without score calibration.

The takeaway

Combining dense semantic vectors and sparse lexical BM25 solves retrieval blind spots.

More data & ai pipelines diagrams