Everything a production stack needs.
Not just a search box — a ranking lab. Tune the blend, rerank the top-k, parse filters from language, and watch the metrics move.
Six things most demos skip
Each one is wired to a real engine and visible in the search UI.
Tunable α blending
Slide keyword ↔ semantic weighting live and watch the ranking shift in real time.
Cross-encoder rerank
Two-stage retrieval: a reranker refines the top candidates for precision where it counts.
Natural-language filters
“laptop under 300” parses the price constraint straight out of the query before retrieval.
Live NDCG / Recall / MRR
Labeled-query metrics overlaid in the UI — the winning path is starred, not asserted.
Secured accounts
Email verification, TOTP 2FA, backup codes, Google OAuth, and rate limiting.
MMR diversity
Suppress near-duplicate results so the top-k spans the catalog instead of one cluster.
Every feature is a control you can touch
Nothing is hidden behind a config file — the whole ranking stack is adjustable from the demo.
Mode tabs
Keyword · Semantic · Hybrid · Rerank · Compare
Switch engines per query, or open Compare to see all three at once.
α slider
Tunable α blending
Drag between keyword and semantic weighting; the ranking and metrics update live.
Rerank toggle
Cross-encoder rerank
Flip on two-stage retrieval to refine the top candidates.
MMR toggle
Diversity
Suppress near-duplicates so the top-k spans the catalog.
Query box
Natural-language filters
Type “under 80” or “over 200” and the price constraint is parsed automatically.
Metrics overlay
NDCG / Recall / MRR
Labeled queries surface live scores, with the winning mode starred.