Context
T-CAIREM — the Temerty Centre for AI Research and Education in Medicine at the University of Toronto — sits on some of the most sensitive data there is: clinical records contributed by hospitals for research. Our principal has led its infrastructure team since 2022. The mandate: give researchers real compute against real data, without the data ever leaving governance.
The problem
Health-data research platforms usually fail in one of two directions. Either they lock data down so hard that researchers export samples and work outside the fence — the worst possible outcome — or they loosen controls until compliance teams shut the whole thing down. The Health Data Nexus had to hold both lines at once, and it had to do it across jurisdictions: deployments in the US, Canada, Korea, and Africa, each with its own rules about where bytes may live.
What we built
The platform is a Django-based infrastructure-as-a-service on Google Cloud: researchers request a project, and the system provisions a data-attached notebook — compute that mounts governed datasets in place. Access is gated by an integrated LMS: complete the required training and certification, and permissions unlock automatically.
Identity is OAuth-based with federated site search across all deployments, so a researcher in Toronto and a collaborator in Seoul see one coherent platform. A distributed event-management module coordinates multi-site datathons — Kaggle-style national challenges run directly on governed medical datasets.
Operations are fully codified: Terraform provisions, GitHub Actions deploys, and an automated unit/integration/load test suite defines the Kubernetes scaling benchmarks the platform is held to.
Key decisions
The hardest call was hybrid cloud. Early-stage researchers get GCP elasticity; hospital workloads with sovereignty requirements run on UofT’s SciNet HPC cluster, inside institutional walls. One platform, two substrates, chosen per dataset — that architecture is what made hospital partners comfortable contributing data at all.
Governance is enforced in the pipeline, not the policy binder: data-egress guardrails are code, reviewed in the same pull requests as features, keeping HIPAA and PHIPA compliance continuous rather than an annual audit scramble.
Outcome
The Nexus now serves a four-region research federation, hosts national datathons, and continues to grow — current work includes deployment onto a sovereign compute grid, an Ontario Health long-term-care AI solution, and a dedicated AI training environment. It is the reference architecture for how we approach every regulated-data platform: compliance designed in, never bolted on.