Tutorials · Chapter D (4/4) · ~10 min
Re-ranking lab
Try it → see it → read → next
**Retrieve wide, re-rank narrow** — hybrid search plus a cross-encoder beats vector-only on mixed queries.
Try tools: Code & developers
Try yourself
Playground
Re-rank retrieval lab
Pick keyword, vector, or hybrid+re-rank for each query — then merge top hits.
Find ticket SKU-8842 in support docs
Recap
What you just did
You matched each query to keyword, vector, or hybrid + re-rank.
Read
Pipeline
candidates = merge(keyword_hits, vector_hits, k=20)
top = cross_encoder_rerank(query, candidates, k=5)