> System.Glossary

Terms and abbreviations.

Definitions for acronyms, abbreviations, and technical terms used across the KARECTL platform, website, and related documentation.

Agentic workflow: A multi-step process in which AI agents plan and carry out tasks using tools and data, within the controls set for them.
AI-conformant: KARECTL's term for a platform on which AI models and agents work under the same identity, access, isolation, audit and approval controls as people. KARECTL is working towards CNCF Kubernetes AI Conformance.
Artificial Intelligence (AI): Computer systems that perform tasks typically requiring human reasoning, such as pattern recognition, decision-making, and natural language processing.

Clinical safety: The process of identifying and managing risks to patients from health IT, including AI, before and during its use.
CNCF Kubernetes AI Conformance (working towards): A Cloud Native Computing Foundation programme defining the capabilities a Kubernetes platform needs to run AI and machine learning workloads reliably.

Data Access Committee (DAC): The committee that oversees approval of research projects and operational AI applications before they can use data on KARECTL.
DataSHIELD: An open-source framework for federated statistical analysis of sensitive data. Researchers run analyses across distributed datasets without raw data leaving each site.

Electronic Health Record (EHR): A digital record of a patient's clinical history maintained by a healthcare provider.

Federated analysis: Analysis run where each dataset is held, so that only aggregate results, not raw data, are shared between organisations.

Governed environment: The controlled space within KARECTL where sensitive data is stored and used. People, AI tools and outputs are subject to approval, access control and review.

Information Governance (IG): The policies, procedures, and controls that govern how personal and sensitive data is managed, accessed, and protected within an organisation.
Infrastructure-agnostic: Designed to run on different infrastructure, such as on-premises hardware or a public cloud, without changing how the platform works.
Infrastructuring: The continuous, intentional development of infrastructure together with the practices that use it, a concept described by Pipek and Wulf (2009).

JupyterHub: A multi-user server for Jupyter notebooks. Provisions interactive coding environments for multiple users from shared infrastructure.

K8TRE: An open-source, Kubernetes-native Trusted Research Environment that LTH co-develops. KARECTL is built on K8TRE.
KARECTL: Lancashire Teaching Hospitals' AI-conformant, Kubernetes-native platform for governed AI and analytics on sensitive data, supporting operational and research use on one architecture. Built on K8TRE.
Keycloak: An open-source identity and access management platform. Handles authentication, authorisation, and single sign-on for applications and services.
Kubernetes: An open-source platform for managing containerised applications. It automates deployment, scaling, and operation of application containers across clusters of machines.

Lancashire Teaching Hospitals (LTH): Lancashire Teaching Hospitals NHS Foundation Trust. The NHS organisation that developed and operates KARECTL.
Large Language Model (LLM): A type of AI model trained on large volumes of text data, capable of understanding and generating natural language.

Multi-Factor Authentication (MFA): An authentication method that requires users to provide two or more verification factors before access is granted.

National Health Service (NHS): The publicly funded health service of the United Kingdom.

Operational AI: AI applications used to run or improve live services, such as follow-up triage or waiting-list prioritisation, as distinct from research.

Public Advisory and Accountability Group (PAAG): A group of members of the public who review how patient data is used on KARECTL, for research and services, and challenge us where needed.

Research output review: The review of research results, including statistical disclosure control, before they are released from the governed environment.
RStudio: An integrated development environment for R.

Scale to zero: Shutting down idle workspaces completely so they use no compute until they are needed again.
Self-hosted inference: Running AI models on an organisation's own infrastructure, so data is not sent to third-party AI services.
Statistical Disclosure Control (SDC): Checks applied to research outputs to reduce the risk that any individual can be identified from them.

Trusted Research Environment (TRE): A secure, governed platform that provides researchers with controlled access to sensitive data. Data does not leave the environment; analysis is performed within it.

vLLM: An open-source library for fast LLM inference and serving. Used in KARECTL to provide local, privacy-preserving model inference without external data transfer.