How CareCompile Works | Multi-Agent Clinical AI, HL7 + FHIR Pipeline

Step 1

Hospital EMR Sends Data

→

Step 2

CareCompile Ingests & Queues

→

Step 3

AI Analyzes Automatically

→

Step 4

Clinicians Review & Act

1

Hospital EMR Sends Clinical Data

MediFlow (the hospital's electronic medical record system) sends data to CareCompile over secure HTTPS connections — automatically, every time something happens with a patient.

MediFlow EMR Hospital System

→

HL7 Messages HTTPS · Port 443

→

FHIR Resources HTTPS · Port 8081

→

CareCompile Receives Logged & Queued

Message types received

ADT — Admissions & Transfers ORU — Lab Results MDM — Clinical Documents DFT — Billing Charges FHIR Patient Records FHIR Culture Results (Blood / Urine) EKG Images 27 total message types per patient

2

AI Analyzes Everything — Automatically

No human triggers required. The moment data arrives, the AI pipeline fires. Here is where the intelligence lives.

Job Queue Prioritized

→

Smart Chunker Context Segmentation

→

5 AI Specialists Running in parallel

→

Deep Clinical Reports Per specialty

→

Executive Summary + Risk Score

Pass 1 — Fast Screening
Five Specialist Models
Llama 3.3 · 70B Parameters · Meta AI
Five clinical specialists powered by Llama 3.3 70B run simultaneously the moment patient data arrives. Each reads the full clinical picture and flags anything concerning in their domain.

        Lab Interpreter
        Pharmacist
        Cardiologist
        Nephrologist
        Internist
      

Pass 2 — Deep Analysis
Full Clinical Narrative
Qwen 3.5 · 397B Parameters · Alibaba AI
One of the world's largest clinical reasoning models. Reads the specialist findings and writes a full, peer-quality clinical report per specialty — the same depth a senior consultant would produce.

        397 Billion Parameters
        Full Narrative
        Cross-specialty Synthesis
      

Pass 3 — Executive Summary

Risk Score + Action Items

CareCompile Expert Engine

Synthesizes all specialist findings into a single executive summary with a clinical risk score, prioritized considerations, and suggested next steps — ready for the attending physician.

Risk Score Priority Flags Treatment Considerations CMO Synthesis Card

Automatic Escalation: When critical lab values are detected (e.g. severe potassium, creatinine, troponin), the system automatically triggers a second AI validation pass — a separate Llama model independently re-reads the case to confirm or escalate the finding before it reaches the clinician.

3

Clinicians Review AI Insights

CareCompile surfaces everything the AI found through six focused views — each designed for a specific clinical workflow. Click any card to see its full technical architecture.

Live Feed

Watch HL7 messages arrive from MediFlow in real time. A live pipeline monitor showing the 5-step automation — Queue → Chunks → AI → Flags → Critical Alerts. Updates every 5 seconds.

View Architecture Diagram

Census

The "God's View" — all admitted patients at a glance, grouped by unit (ICU → ED → Med-Surg). AI-generated alert cards per patient. Dashboard counts, billing pipeline, and ward-level summaries. Auto-refreshes every 60 seconds.

View Architecture Diagram

Visit Analysis

Deep dive into a single patient visit. Six tabs: Overview, Diagnostics, Documentation, AI Analysis, Actions, Expert. The full output of all three AI passes lives here — readable by the attending in under 2 minutes.

View Architecture Diagram

Flags & Alerts

AI-generated clinical warnings surfaced from all specialist analyses. Physicians can acknowledge, snooze, or escalate each flag. Eight filter categories — Critical, Drug Interactions, Lab Trends, and more.

View Architecture Diagram