Results Dossier

Built for the work
that can't afford to be wrong.

We ship agentic AI into regulated, high-stakes workflows — tax advisory, estate law, municipal finance, benefits underwriting, field operations, healthcare, consumer goods, and private markets. Pick a vertical to see the deliverable, what it replaced, and the outcome it drove.

Healthcare

From Medallion to Knowledge Graph

We rebuilt the hospital gold layer for MRF and payer Transparency in Coverage price data — replacing a pile of dbt marts, where every new question needed a new mart and no join key survived across sources, with a knowledge graph that lives as a logical layer over the same BigQuery tables.

Gold layer

MedallionKnowledge Graph

  • Entity Resolution

    NPPES is the provider spine; blocking, probabilistic scoring, and transitive clustering do the rest, with embeddings for messy facility names. Versioned cluster IDs mean merges never break downstream keys.

  • Reconciliation

    MRF and TiC disagree often. We never overwrite — conflicting values become sibling assertions, and a precedence policy derives a consensus view. Divergence becomes a tracked signal, not a COALESCE.

  • The Ontology

    We reused rather than invented. FHIR R4 supplies the core classes; SKOS carries CPT, MS-DRG, and NDC. Rates become reified assertion nodes — a price is an n-ary relationship, not an edge property — each carrying PROV-O provenance.

Outcome

Questions became traversals instead of marts, and the gold backlog dissolved. The graph lives as a logical layer over the same BigQuery tables — no second store, no ID handoff between engines.

Contact

Let's Talk

Tell us what you're working on. We'll let you know if we can help — and if we can't, we'll point you in the right direction.

Response time: Within 24 hours

Email: Nick@gettom.io