Platform

Architecture, ontology, and execution engine

Six tiers. Seven domains. Twenty-plus subsystems. One ontology. One execution engine. Edge-first, sovereignty-respecting, deployable where connectivity is a variable.

Core Design Principles

01Decentralization-FirstNo single point of failure. Every hospital runs a complete local cluster. If the cloud disappears, the institution keeps operating at full intelligence on edge hardware.

02Ontology-DrivenA living, computable model of institutional reality: every patient, provider, resource, event, and relationship. The single source of truth.

03Insight-to-ExecutionNovaOR doesn't stop at recommendations. Intelligence converts into automated workflows, triggered actions, and measurable outcomes — always with human oversight.

04Vendor AgnosticismBuilt on Kubernetes, ONNX, FHIR, MQTT, and open standards. Runs on any cloud or bare metal. Zero vendor lock-in by architecture.

05Clinical Safety FirstAI never overrides a physician. Every clinical recommendation includes explainability, confidence scores, and source citations.

06Privacy-NativeGDPR, HIPAA, Morocco CNDP compliant by architecture. Federated learning ensures patient data never leaves the hospital.

07Offline ResilienceFull operational capability in connectivity-degraded environments. USB-bootable emergency mode for catastrophic infrastructure failures.

08Progressive AdoptionStart with core scheduling and monitoring free. Advanced AI and specialized domains unlock progressively. Zero disruption to existing workflows.

Interactive Architecture Model

Click a domain node or layer to explore. Hover for details.

nova-os · architecture.iso · interactivebuild 0.4.1 · edge · signed

The Six-Tier Architecture

NovaOR operates across six distinct tiers, each with defined responsibilities, failure boundaries, and communication protocols. The critical invariant: Tier 2 (hospital edge) is the source of truth and can serve 100% of workloads when all other tiers are unreachable.

Global Intelligence

Multi-region, multi-cloud

Federated learning aggregation, global model registry, cross-hospital benchmarks, NovaMarketplace

Tiers 1–5 operate fully autonomously. Sync resumes on reconnect.

Cloud (Hospital Group)

Cloud region per hospital group

Model training (MLflow + Kubeflow), group-level analytics, cross-hospital federation hub

Edge tiers operate autonomously. Sync resumes automatically.

Edge (Per Hospital)

On-premise K3s cluster

All 20+ domain subsystems, local ONNX inference, local databases, Kafka event bus, NovaCard auth

Fully self-sufficient. 100% of hospital operations offline indefinitely.

Department / Ward

Department-level servers

Department dashboards, local caching, ward-level alert routing, scheduling views

Falls back to Tier 2. No data loss (event-sourced).

Device

Tablets, phones, watches

Staff interfaces, patient apps, NovaCard NFC readers, biometric terminals

Individual device failure; others continue.

Sensor / IoT

Bedside monitors, RFID, wristbands

Vital sign streams, equipment tracking, room environment, medication dispensing

Graceful degradation. Missing data flagged.

The NovaOR Ontology — Modeling Healthcare Reality

Legacy hospital systems store data in isolated silos. NovaOR begins with a fundamentally different premise: to drive intelligent execution, you must first build a living, computable model of reality.

The NovaOR Ontology is that model — a continuously updated knowledge graph representing every entity, relationship, event, and state in the healthcare institution. Stored in Neo4j and updated in real time by every subsystem.

The Five Entity Classes

Person

Patient, Physician, Nurse, Resident, Administrator

Resource

OR Room, Bed, Ventilator, MRI Scanner, Surgical Kit, Drug

Process

Surgery, Consultation, Lab Test, Drug Administration, Training

Event

Vital Anomaly, Equipment Failure, Medication Error, Code Blue

Knowledge

Clinical Protocol, Drug Interaction, Diagnosis, Guideline

The Insight-to-Execution Pipeline

What separates NovaOR from every analytics tool on the market. When a sepsis risk is detected, NovaOR does not display a dashboard tile — it generates a clinical alert, notifies the physician, pre-stages the sepsis bundle, adjusts nursing priority, increases vital sign polling, and logs every step for audit.

01ObserveEvery sensor, system, and human interaction continuously feeds real-time data into the NovaOR event fabric.

ExampleBedside monitor streams SpO₂ at 88% for 90 seconds.

02UnderstandThe Ontology Engine maps observations to entities and updates the knowledge graph. ML models analyze patterns and detect anomalies.

ExamplePatient entity updated: SpO₂ below threshold. Risk model fires: 73% probability of early respiratory failure.

03DecideThe Intelligence Layer generates recommended actions with confidence scores, rationale, and alternatives. Clinical decisions require human approval.

ExampleAlert generated: 'Consider O₂ supplementation + pulmonology consult.' Nurse notified via smartwatch.

04ExecuteApproved actions trigger automated workflows across the relevant subsystems: schedule changes, resource allocations, communications, billing codes.

ExampleOxygen request logged. Pulmonologist scheduled. Event recorded in audit trail.

AI Ethics & Governance Framework

Five non-negotiable pillars, each with specific technical enforcement mechanisms. Not aspirational — architecturally enforced.

Interoperability Layer

NovaOR connects to any hospital's existing ecosystem without requiring a rip-and-replace. For hospitals with no existing EHR, NovaOR operates as a lightweight EHR itself with a full transition pathway.

Inbound: Patient data flows from the hospital's EHR into NovaOR via FHIR subscriptions or HL7 v2 feeds. Outbound: AI recommendations, scheduling updates, and care plans flow back as FHIR resources.

HL7 FHIR R4Bidirectional patient data exchange with any modern EHR

HL7 v2.xLegacy EHR integration for hospitals not yet on FHIR

DICOM + DICOMwebMedical imaging ingestion from PACS; structured AI reports

IHE ProfilesCross-system interoperability workflows (XDS.b, PIX, ATNA)

SMART on FHIRThird-party app integration via standardized launch protocol

SNOMED CT / ICD-10/11Clinical terminology for diagnoses, coding, and reporting

LOINCLaboratory observation codes for lab result interpretation

RxNormMedication naming for drug interaction checking