Fuzzy search and data
Fuzzy matching is not a convenience feature. It is the core commercial wedge.
Retailers lose sales and users lose trust when sauvaj, bleeu, Dior edp,
or br540 extrait return the wrong item or no item.
Matching rules
The resolver must consider:
- fragrance name
- brand and parent house aliases
- concentration
- flanker family
- year and reformulation signals
- perfumer
- note pyramid and accord profile
- country and distributor aliases
- SKU/product-title noise
- confidence tier and source quality
Low confidence must ask the user or merchant to choose from candidates. Silent wrong snapping is worse than asking.
Sellable formats
- API endpoint for developers.
- Hosted search widget for retailers.
- Shopify storefront search extension.
- Batch catalogue cleanup.
- Feed enrichment export.
- Retail staff iPad search.
- Scent Profile Connect matching at checkout.
Accuracy workflow
- Parse raw title or query.
- Generate candidate identities.
- Score candidates with deterministic features first.
- Use embeddings as a secondary signal.
- Return candidates with evidence, not just a single id.
- Store corrections so the resolver improves.