FHIR Data Enrichment, Curation and Data Products 101 - Part 1
Detailing your data
The Autonomous FHIR Data Factory
Part I: From Chaos to Foundation — Building the Trust Layer for AI in Healthcare
Without trusted data, even the best AI is just guessing.
We’re in a healthcare AI boom but too many projects are failing silently. Why? Because they’re built on data foundations that were never designed for automation, let alone intelligence.
This is the first installment of a four-part series outlining a new architectural paradigm: The Autonomous FHIR Data Factory an agent-driven, standards-based, AI-ready platform for transforming raw, messy healthcare data into high-fidelity, decision-ready assets.
This post focuses on the first step: laying the foundation. Before AI agents can curate, enrich, and generate insights, we must solve for trust, structure, and usability at scale.
🚨 Interoperability Is Not the Goal. Usability Is.
Most healthcare systems are drowning in data silos. Even with APIs and health information exchanges, we remain in an era of interoperability theater. Systems can technically exchange data but clinicians and algorithms alike don’t trust it, can’t use it, and often ignore it.
We must shift from exchanging data to producing usable, enriched, and governed health information.
At the core of this transformation is FHIR not just as a compliance framework, but as the scaffolding for building a new kind of health data factory.
🧱 The Architecture: From Raw to Product
Before AI or analytics, we need a trusted pipeline. Here’s what it looks like:
This layered model keeps raw data intact (audit-ability), supports targeted enrichment and curation, and enables plug-and-play data product generation.
🔍 Start With Discovery, Not Code
Before ingesting any data, we begin with discovery and profiling.
Identify a real-world use case (e.g. unified medication list)
Profile source systems for structure, quality, and semantics
Document mappings, terminology conflicts, and quality rules
🎯 This creates not just mappings but machine-readable blueprints for automation.
📏 Scoring Trust: The Data Quality Matrix
We don’t guess whether data is “clean enough.” We score it—objectively.
Each source system is measured across six dimensions (accuracy, completeness, conformity, etc.) with defined thresholds and automated actions on failure.
This becomes the contract between upstream systems and the data factory. And it’s 100% automatable.
🔁 The Raw FHIR Layer: Preserve First, Perfect Later
Rather than cleanse everything up front (and lose information), we map to a Raw FHIR Layer:
Source formats → FHIR resources
Validate structural conformance
Log every issue, but don’t block load
Think of it as a read-only ledger of imperfect reality the starting point for all future automation.
🧪 Post-Load: Semantic Validation at Scale
Now that the data is in FHIR, we run higher-order checks against:
USCDI Data Classes (what must be exchanged)
US Core IG (how it must be represented in FHIR)
Tools like Inferno validate for “must support” fields, value set compliance (e.g. LOINC, SNOMED), and reference integrity.
The result? A Gap Analysis our prioritized to-do list for enrichment and curation.
This workflow shows how agents collaborate: each specializing in ingestion, assessment, enrichment, and beyond.
🗺️ Strategic Execution: From Use Case to Delivery
Don’t start with a massive platform build. Start with a use case. Map, load, score, and produce.
This timeline shows how organizations can go from idea to trusted data product with focused scope and measurable progress.
This modular architecture is standards-driven, auditable via FHIR Provenance, and designed to scale with open tooling.
💬 Why It Matters
The Autonomous FHIR Data Factory is more than a technology vision it’s a strategic operating model.
It’s how we move from the chaos of legacy systems to an ecosystem where LLMs can reason, clinicians can trust, and innovation can scale.
In the next post, we’ll cover:
🔬 Semantic Enrichment using LLMs and terminology services
🧠 Curation logic for golden records
📦 Packaging into smart, reusable FHIR data products
🙌 Let’s Build It Together
If you're building in this space let’s talk.
If you disagree challenge it.
If you’re inspired share it.
🔁 Subscribe, 🧵 Comment, and 💥 Join the movement.
Written by Eugene Vestel, Founder of FHIR IQ | Building the future of trusted healthcare data, one agent at a time.