genix

FHIR R4 + HL7 v2 in Production Medical AI: Building Bodh

How to architect medical AI systems that interoperate with legacy health systems while maintaining modern safety constraints.

Why Healthcare Needs Two Standards

Key FHIR R4 Resources

1. Patient - Demographic and personal information
2. Observation - Clinical measurements and test results
3. Condition - Diagnoses and clinical problems
4. Medication - Drug information and prescriptions
5. CarePlan - Care coordination and treatment plans

FHIR R4 (Fast Healthcare Interoperability Resources)

• Modern, REST-based, JSON/XML
• Used by newer EHRs (Epic, Cerner 2020+)

HL7 v2

• Legacy standard (1989–present)
• Flat, pipe-delimited messages
• Used by 90% of hospitals globally

You need to support BOTH because your healthcare customers use both.

Bodh: Medical AI Panel Architecture

Bodh is an open-source medical AI panel on Microsoft’s MARA with:

• 7 role-specialized agents: intake, supervisor, questioner, test_planner, cost_guardian, diagnostician, reasoning_verifier
• FHIR R4 schemas for EHR interoperability
• HL7 v2 message parsing for legacy system integration
• Cost-aware diagnostic budgets that agents enforce
• SD-Bench evaluation metrics for multi-turn diagnostic reasoning

Cost-Aware Diagnostic Budgets

In healthcare, every test costs money. The cost_guardian agent enforces a budget:

Patient presents with cough
├─ Budget: $500
├─ Intake: fever, cough, shortness of breath
├─ Tests: Chest X-ray ($200), CBC ($50), Culture ($100)
├─ Cost guardian: "Budget allows all three. Proceed."
└─ Result: Pneumonia (confirmed by imaging + labs)

HIPAA Compliance & Data Privacy

1. Encryption in transit: TLS 1.3
2. Encryption at rest: AES-256-GCM
3. Audit logging: Every data access logged
4. Consent management: Track which agents can access what
5. Data residency: Keep patient data in-country per regulations

---

Bodh is open source: github.com/c2siorg/bodh