Skip to main content
Product

fi-fhir

fi-fhir handles healthcare ingestion, transformation, and delivery in the FlexInfer stack. It turns legacy clinical formats into semantic events and routable workflows, with AI-assisted development paths that can run against private FlexInfer models instead of moving healthcare data into a shared SaaS control plane.

Read fi-fhir docsOpen playgroundRead case studySource code

fi-fhir, on the record

Read from the Mills factory ledger, the synced docs, and the edilint build this site ships.

Reading live figures…
  • Merged by Mills
  • Docs
  • Playground
Healthcare ETL

From legacy feeds to explainable events

Healthcare integration work usually fails in the gaps between source quirks, validation, terminology, and operations. fi-fhir makes those boundaries explicit so they can be tested with repeatable samples and explicit contracts.

  1. Configure

    Source Profiles

    Every feed carries its own configuration, mappings, validation expectations, and routing behavior, with vendor starter templates for Epic, Cerner, Meditech, and Allscripts that fork per interface.

  2. Ingest

    Format-agnostic ingestion

    HL7v2, CSV, EDI X12, and CDA/CCDA inputs become semantic events that workflows can validate, transform, and route.

  3. Route

    Workflow routing

    YAML workflow definitions with CEL filters route events to FHIR APIs, webhooks, databases, Kafka, Redis Streams, Google Cloud Pub/Sub, or validation-only test paths, with a GraphQL API (Beta in the status matrix) and a TypeScript SDK around the platform.

  4. Deliver

    FHIR and terminology output

    FHIR R4 transaction delivery covers admission, lab, and six additional clinical event families, with Patient/Encounter references resolved. Pinned structural checks and live-server tests define coverage; official conformance validation remains open.

Integration engine

From parsing library to feed runtime

The engine connects authenticated ingress, profile selection, parsing, durable receipts, persistence, and workflow delivery. fi-fhir remains pre-1.0; the case study records implemented paths and the tests and release work still outstanding.

Component statusSynced docs · Sep 20

  • 33Production
  • 11Beta
  • 9Alpha
  • 2Planned

Components by maturity, as the repo’s status matrix states them; its latest dated update is Sep 20, 2026.

  1. Ingress

    Authenticated ingress

    Authenticated HTTP, MLLP, and S3/SFTP batch paths feed one durable processor. HTTP HL7 ingress checks service, source, and tenant identity before accepting a submission.

  2. Acceptance

    Durable acceptance

    Atomic durable submission gives every accepted message a receipt, idempotent duplicate handling, and restart-safe persistence before delivery.

  3. Testing

    Deterministic preview

    Preview mode runs the same parsing and routing semantics as production without persisting or delivering, so feed tests exercise the real execution path.

  4. Governance

    Versioned, governed artifacts

    Immutable profile and workflow revisions, typed secret references, and PHI-aware data classification keep integration changes auditable.

AI-assisted development

Keep assistance close to the data boundary

AI assistance supports specific tasks: help engineers explain warnings, draft workflows, and search terminology while keeping private healthcare context aligned with the runtime and governance boundary.

  1. Explain

    Warning explanations

    AI-assisted explanations turn parse warnings and opaque healthcare field names into reviewable reasoning instead of tribal knowledge.

  2. Draft

    Workflow generation

    Natural-language workflow drafting and explain-back run against live LLM operations in the Mapping Studio, for operators who know the integration behavior but not the DSL shape yet.

  3. Search

    Semantic terminology search

    Meaning-based terminology search helps bridge local code systems, LOINC, RxNorm, and downstream FHIR expectations.

  4. Score

    Entity extraction and quality scoring

    Clinical entity extraction and data-quality scoring run against any OpenAI-compatible endpoint, behind runtime capability checks.

Stack relationship

Connect healthcare workflows to private models

FlexInfer serves the models behind explanations and workflow drafting. Loom governs MCP, context, and agent access. MentatLab can orchestrate multi-step integration workflows as DAGs. fi-fhir supplies the parsing, mapping, validation, and delivery contracts. Data access and model permissions still need to be configured for each deployment.

fi-fhir | FlexInfer