Concepts

The building blocks behind Vitals — the data-engineering and healthcare ideas the pipeline is made of, explained plainly. If a term on another page is unfamiliar, it's defined here.

Building blocks

• Healthcare data: FHIR · PHI · OMOP — the three ideas that make this a health-data project, and how they fit together (FHIR in → de-identify → OMOP out).
• Glossary — every term used across the site, one line each.

The shape of the whole thing, in five sentences

1. Medallion — data flows through three quality tiers: bronze (raw), silver (cleaned & de-identified), gold (ready to consume).
2. Sources arrive as FHIR (the clinical-data exchange standard), plus claims, patient-reported outcomes, wearables, and notes.
3. Silver is the trust boundary — PHI (protected health information) is removed here, codes are standardized, and data quality is enforced.
4. Gold is served three ways because analytics, ML, and LLM/RAG each need a different shape: dbt analytics marts (conformed to OMOP), a feature store, and a vector index.
5. It runs locally on DuckDB for reproducibility, with Databricks/Delta as the deployment target and Spark for scale.

See Architecture for the diagram and Results for what the pipeline actually produces.