> System.Architecture
Designed in layers.
Governed at every level.
KARECTL is built in four layers on a Kubernetes-native foundation, with security, governance and audit enforced at every layer, for operational and research workloads alike.
Four layers, from Infrastructure through Platform Foundation and Core Services to Applications. Each band lists its capabilities. Security, governance and audit apply at every layer.
01 / Infrastructure
- Kubernetes
- On-premises
- Microsoft Azure
02 / Platform Foundation
- Identity & access (Keycloak)
- Network isolation
- Certificates & secrets
- Encrypted storage
03 / Core Services
- Observability
- AI-assisted remediation
- Provisioning
- Data & messaging
- Security as code
- AI inference (vLLM)
- Workflow orchestration
04 / Applications
- Operational AI applications
- Research workspaces
- Federated analysis (DataSHIELD)
- Agentic research tools (in development)
> Layer Breakdown
01 / Infrastructure
Kubernetes on-premises and on Microsoft Azure. KARECTL is designed to be infrastructure-agnostic, so organisations can choose where it runs.
02 / Platform Foundation
Shared services abstracted from the underlying infrastructure: identity and access management with multi-factor authentication (Keycloak), network routing and isolation, certificate and secrets management, and encrypted storage.
03 / Core Services
Observability and AI-assisted remediation; project and service provisioning; data storage and messaging; security and compliance defined as code, covering policy enforcement, runtime threat detection and configuration scanning; and AI services, including self-hosted model inference (vLLM), LLM observability, workflow orchestration and event-driven scaling.
04 / Applications
Operational AI applications, such as follow-up triage and waiting-list prioritisation, run in their own governed environments. Researchers work in isolated project workspaces with R and Python, JupyterHub notebooks, RStudio, shared code repositories, remote desktops and DataSHIELD for federated analysis. Agentic research tools are in development.