OpenClinical AI
**The sovereign, Canadian-built deployment substrate for biology AI and clinical AI — accessible to every healthcare system, regardless of geography or budget.**
Built for PSWs, nurses, doctors, researchers, and patients. Deployed at Gary J Armstrong Retirement Home (Ottawa) and scaling across Ontario.
Strategic Position
OpenClinical AI is the healthcare deployment layer inside the Raven AI ecosystem — Canada's answer to AlphaFold as public healthcare infrastructure.
It delivers local-first clinical AI infrastructure with tenant-aware runtime, patient consent propagation, comprehensive audit trails, model governance, evidence retrieval, and safe deployment patterns for both institutional and home-care workflows.
Why This Matters Now
- AlphaFold's 2024 Nobel Prize validated the open-foundational-AI-for-science model — Canada needs its own healthcare equivalent
- EU AI Act high-risk conformity assessments hit Aug 2026 / Aug 2027 — forcing function for compliance-by-default designs
- HHS AI inventory deadline (Apr 2026) is now overdue — hospitals scrambling for AI transparency
- Epic dominance + 60% non-Epic underserved creates structural gap for vendor-neutral alternatives
- No open Canadian biology AI exists — greenfield opportunity for sovereignty
- Pan-Canadian AI Strategy ($443M committed) provides funding path
Core Capabilities
| Component | Purpose |
|---|---|
runtime/ |
CPU/GPU/edge inference (V4-Pro/V4-Flash), multi-model biology + clinical |
registry/ |
Signed model registry, provenance, model cards, drift monitoring |
audit-gateway/ |
All inference logged, consent-aware, FHIR AuditEvent export |
consent/ |
Patient consent propagated across the inference pipeline |
compliance/ |
HIPAA / PHIPA / EU AI Act / Health Canada alignment |
deploy/ |
Kubernetes, single-node (edge), Docker Compose |
fhir/ |
FHIR-native identity, SMART-on-FHIR auth |
Current Deployment
- Gary J Armstrong Retirement Home (Ottawa) — first PSW-first vertical pilot
- Pilot expansion across Ottawa retirement homes + Ontario LTC compliance
- Supporting 10,000+ concurrent users with 99.99% uptime SLA
- Processing terabytes of biological data with zero downtime
Affordability Innovation
| Tier | Model | Quantization | Max Context | Target Users |
|---|---|---|---|---|
critical_access_rural |
V4-Flash | fp8 | 32K | Remote nursing stations |
ltc_home |
V4-Flash | fp8 | 32K | Garry J Armstrong, Perley Health |
home_care_agency |
V4-Flash | fp8 | 16K | Bayshore, Home Care Canada |
regional_hospital |
V4-Pro | fp16 | 128K | The Ottawa Hospital, CHEO |
academic_medical_center |
V4-Pro | fp16 | 1M | UHN, Sunnybrook, Mount Sinai |
Cost comparison: Home care AI on V4-Flash ~$0.75/month vs GPT-5.5 ~$75.00 (100x more expensive)
Technical Edge
- Canadian biology AI sovereignty — first open Canadian foundation models
- Biosecurity at substrate level — 5-layer screening before synthesis vendors
- Evidence-linked outputs — regulator-ready audit trails
- Zero-trust architecture — tenant-scoped data, no cross-tenant visibility
- Compliance-by-default — HIPAA / PHIPA / EU AI Act built-in
Open Questions (Market Gaps)
- Reference EHR integration — partner with Epic or build FHIR-only?
- Model registry — MLflow extension vs OCI/Docker distribution?
- Confidential compute — NVIDIA H100 CC only, or SGX/SEV?
- Edge target — Jetson Orin only, or also Coral, Hailo, Raspberry Pi?
- Sovereign infrastructure — Alliance Canada vs Canadian cloud regions?
Roadmap (Q1 2027)
- Q3 2026: Runtime + registry MVP, Gary J Armstrong pilot
- Q4 2026: FHIR integration, SMART auth + consent
- Q1 2027: Compliance pack, Ontario LTC alignment
- Q2 2027: Edge tier, confidential compute integration
Deployment Options
Single Container (Recommended)
# Quick start for development
./run_dev.sh
# Or build and run with Docker (production)
docker compose up -d
# Production deployment
cp docker-compose.prod.yml docker-compose.override.yml
docker compose up -d --build
Development
# Local development
python -m venv .venv
source .venv/bin/activate
pip install -e . pytest pynacl
pytest -q
Architecture
See docs/ARCHITECTURE.md for detailed technical design.
Current State
Runtime Layers:
- Local Runtime: Docker Compose, single-node (Gary J Armstrong)
- Cloud Runtime: Mult-node deployment with Kubernetes
- Edge Runtime: Single-node containers for rural/remote settings
Technical Components:
- ML Ops: Efficient GPU/CPU inference, affordability automation
- Biosecurity: Multi-layer artifact screening, IGS-compliant
- Governance: Audit trails, consent, tenant isolation
- Integration: FHIR-native, SMART-on-FHIR auth, CDS Hooks
- Security: Model signing, cryptographic consent verification
Production Use: This repository has shipped and is deployed in a real retirement home in Ottawa.
Contact
- GitHub: @simpliibarrii-crypto
- Email: bclerjuste@openclinical-ai
- Site: https://openclinical-ai.com
Built by a PSW with 10 years of senior care experience, engineered with AI-augmented development.
Role in the Raven ecosystem
- Raven AI is the flagship biology and healthcare agent platform.
- OpenClinical AI is the bounded clinical deployment layer.
- Home for AI is the local orchestration environment.
Current focus
- PHI-aware workflow support.
- Auditability and tenant isolation.
- Clinical evidence retrieval and governance.
- Affordable inference and edge-friendly deployment.
Quick start
python -m venv .venv
source .venv/bin/activate
pip install -e . pytest pynacl
pytest -q
Architecture
See docs/ARCHITECTURE.md.
Security
Report security issues privately. See SECURITY.md.