Fhirlace laces an existing healthcare database into the FHIR fabric — without moving a single row. Connect it, review the mapping, and serve a full FHIR API from the data you already have.
Fhirlace runs the model once, at setup — never at request time. What serves your data is a deterministic engine reading one reviewed file, so the same row always produces the same resource.
Fhirlace introspects the schema, profiles every column for data-quality caveats, and infers relationships. The AI proposes a FHIR mapping — with a confidence score on each field.
Low-confidence fields are flagged. Nothing serves live data until a reviewer approves it. Automated mapping isn't accurate enough to trust blindly — so we don't.
◈ clinical review requiredA stateless engine answers FHIR REST calls straight from your database, and writes FHIR back into the existing schema. No copy, no drift, fully auditable.
Built from the ground up as bidirectional, terminology-aware, and auditable — the parts a naïve converter skips.
PostgreSQL, MySQL/MariaDB, SQL Server, Oracle, SQLite — introspected through one universal connector.
Read as FHIR, write FHIR back. Every transform carries an inverse, and non-reversible fields are read-only by construction.
Bare codes become validated CodeableConcepts — LOINC, ICD-10, SNOMED — with locale-aware display and cross-system translation.
Read, search, create, update, delete, Bundle transactions, _include and CapabilityStatement.
Dialect-aware upserts into the existing schema, with read-before-write conflict detection so concurrent edits never silently clobber.
Fhirlace fingerprints the schema and tells you the moment it changes — so the FHIR surface never silently falls out of sync.
Storing J06.9 and emitting "J06.9" is structurally valid and clinically useless. Fhirlace resolves codes to full CodeableConcepts — and can serve them in the system an Implementation Guide demands, losslessly translating back on write.
Automated schema-to-FHIR mapping tops out around 70% accuracy and hallucinates plausible-but-wrong mappings. So Fhirlace treats human review as a first-class stage — not a formality — with a promotion gate no mapping can skip.
low-confidence field detected → blocked until sign-off
A connection string is all it needs to introspect, profile, and draft a mapping. You review; it serves.