Intelligence infrastructure for wine applications
Sommelier Engine resolves a wine, infers what it doesn't directly know, and composes a coherent, explainable profile from a versioned knowledge graph — served through one stable API. It's an engine, not a database.
{
"identity": "Barolo",
"vintage_year": 2016
}
→ {
"snapshot_id": "sha256:0e0e231f…",
"handle": "[email protected]/1.1",
"version": { "research":"1.1","composition":"1.1","engine":"0.1","api":"v1" },
"profile": { "descriptors": [ { "descriptor":"tar","confidence":0.9,
"dominant_mechanism":"varietal_aroma" } ], "axes": { … } }
}Four independent version axes: research · composition · engine · api. Reproducible forever.
Same request + same version ⇒ byte-identical snapshot_id. Proven across environments.
Not a black box — every answer resolves to mechanisms and evidence.
A stable, read-only contract. Build products on knowledge, not scraping.
A real technology stack
Restaurant → Resolver → Inference Engine → Knowledge Graph → Composition → Snapshot → API → Application.
Identify & normalize any wine reference into a canonical identity.
Derive knowledge for unknown or partial wines via controlled backoff + priors. The moat.
38k wines distilled into 125 identities × 40 grape baselines, styles, vintage, bottle evolution.
Mechanism-owned, signed, de-duplicated composition — coherent profiles, never double-counted.
Every descriptor carries a reconstructible causal chain to its evidence.
Build on wine knowledge, not scraping.
Reproducible, explainable, versioned. One API.