HL7AI Studio · Product

HL7AI Integration Agent

An HL7 hub that routes your messages to your integration engines (Mirth, Rhapsody…) and applications — with a local AI that, per channel and per message type, fills documents, summarizes and extracts information. Every generated output is clearly marked as AI.

V2.x
HL7 messaging
V3
HL7 version 3
FHIR
R4 · R5
DICOM
imaging
HL7 hub

A central hub that routes to all your engines

Receive once, route everywhere: the Integration Agent dispatches each message to the right engines (Mirth Connect, Rhapsody…) and applications, by channel and message type.

Messages routed / day
1,2 M
▲ 4,2 %
Active channels
38
all UP
Connected engines
6
Mirth +5
AI tasks / day
84 200
▲ 11 %

Routing: sources → channels → engines

Daily volume · thickness ∝ message count
LAB → cRES: 300RAD → cIMG: 180ADM → cADT: 260PHA → cORD: 120cRES → MIRTH: 140cRES → FHIR: 100cRES → BI: 60cIMG → RHAP: 110cIMG → FHIR: 70cADT → MIRTH: 120cADT → DPI: 90cADT → BI: 50cORD → MIRTH: 70cORD → DPI: 50LabRadiologyAdmissionPharmacyResults (ORU)Imaging (ORM)ADTOrders (RDE)MirthRhapsodyFHIR serverEHRBI warehouse
A message's journey

Receive → AI → route, in milliseconds

1

Receive

MLLP listening, automatic ACK (V2.x, V3, FHIR, DICOM).

2

Validation

Conformance, identifier normalization (NAM/NIM/IPM).

3

Channel routing

Rules by facility, source system and message type.

4

AI by type

Fill, summarize, extract, classify — by message type.

5

Delivery

Mirth, Rhapsody, FHIR server, EHR, BI warehouse.

Routing by channel & type

Each channel has its destinations, transform and AI task

Channel Message type Destinations Transform IA
Laboratory ORU^R01 MirthFHIRBI V2 → FHIR DiagnosticReport Critical-result summary
Radiology ORM^O01 ORU^R01 RhapsodyFHIR V2 → FHIR ImagingStudy Report summary
Admission ADT^A01/A02/A03 MirthDPIBI PID/PV1 normalization Patient-duplicate detection
Pharmacy RDE^O11 MirthDPI V2 → FHIR MedicationRequest Interaction check
Microbiology ORU^R01 MirthFHIR Antibiogram structuring Extraction & coding (LOINC/SNOMED)
Documents MDM^T02 DPIArchive V2 → CDA / FHIR DocumentReference Document fill & summary
AI by channel & type

A specialized local AI for each flow

You ask HL7AI to fill a document, summarize a stay or extract data — the task is applied automatically to the right channel and message type.

AI task distribution

100 tasks
  • Fill documents 34%
  • Summarize 28%
  • Extract & classify 22%
  • Normalize & code 16%

Fill documents

Generates reports, CDA and FHIR DocumentReference from messages.

Summarize

Summary of a stay, a chart or a batch of results.

Extract & classify

Spots critical values, codes to LOINC/SNOMED, categorizes.

Normalize

Reconciles identifiers and terminologies across standards.

Concrete example

Fill a document from an HL7 message

A lab result (ORU^R01) arrives; the AI drafts the report, clearly tagged "AI-generated".

MSH|^~\&|LIS|CHUM|DPI|CHUM|20260628113000||ORU^R01|MSG00231|P|2.5.1 PID|1||TRJE82031517^^^RAMQ||TREMBLAY^JEAN||19580312|M PV1|1|I|SOINS-INT^12^A|||||7654^CARDIN^MARIE OBR|1||LAB-99812|85354-9^Bilan lipidique^LN|||20260628 OBX|1|NM|2093-3^Cholestérol total^LN||6.8|mmol/L|<5.2|H|||F OBX|2|NM|2085-9^HDL^LN||0.9|mmol/L|>1.0|L|||F OBX|3|NM|2089-1^LDL^LN||4.6|mmol/L|<3.4|H|||F
IA

Report — Lipid panel

AI-generated

TREMBLAY, Jean · NAM TRJE 8203 1517 · ICU

Abnormal lipid panel suggesting high cardiovascular risk: elevated total cholesterol (6.8 mmol/L), high LDL (4.6 mmol/L) and low HDL (0.9 mmol/L).

Suggestion: consider a statin and reassess at 3 months.

model hl7ai-local confidence 0.93 2026-06-28 11:30 sources OBX 1-3 to be validated by a clinician
Concrete example

Summarize a stay from multiple messages

The AI aggregates a patient's messages (joined on the key) and produces a traceable summary.

Stay messages

12 HL7 messages · 36 h · NAM TRJE 8203 1517
08:12
ADT^A01 ER admission
08:47
ORU^R01 Troponine 0,92 µg/L
09:05
ORU^R01 Abnormal ECG
09:40
ORM^O01 Angiography
14:00
ADT^A02 Transfer to ICU
IA

Stay summary

AI-generated

Patient admitted to ER for chest pain, then transferred to ICU.

Elevated troponin (0.92 µg/L) and abnormal ECG suggesting acute coronary syndrome; angiography performed.

Plan: continuous monitoring, dual antiplatelet therapy.

model hl7ai-local confidence 0.90 sources 12 msg to be validated by a clinician
AI governance

AI-generated content is always identified

No ambiguity between source data and AI output: each result carries its provenance and stays under human control.

"AI-generated" label

A visible badge marks every AI output in the UI and documents.

Full provenance

Local model, version, timestamp, confidence score and source HL7 segments.

Human validation

Mandatory "to validate" status before clinical delivery (human-in-the-loop).

Traceability

Every output links back to the exact HL7 messages and fields that produced it.

100% local AI

Models run on-premise; no patient data is sent to the cloud.

Audit log

Every AI task is logged (who, what, when, on which source).

Even more powerful with FHIR BI

Transform and route with the Integration Agent, then analyze everything in FHIR BI.

An unhandled error has occurred. Reload 🗙

Rejoining the server...

Rejoin failed... trying again in seconds.

Failed to rejoin.
Please retry or reload the page.

The session has been paused by the server.

Failed to resume the session.
Please retry or reload the page.