Blogs, definitions and sessions on metadata enrichment, data and AI sovereignty, and what it actually takes to make a data estate usable by people and AI agents. Written by the team behind find.data.gov.scot.
Featured blog
Context engineering has become the defining discipline of applied AI, but most of the effort goes into prompts and retrieval, and almost none into the context that already exists in an organisation's metadata. This piece argues that well-structured metadata is the cheapest, most underused lever for making AI systems reliable.
What metadata standards are, why DCAT3, Dublin Core, Schema.org and INSPIRE exist alongside each other, and how to choose between them without locking yourself in.
Jul 2025 · 8 minRead → BlogRaw numbers and variables rarely tell the full story. How AI can generate the descriptive context that makes datasets discoverable, and where human review still matters.
Jan 2025 · 6 minRead → BlogData everywhere, but not easy to find. The story of unifying 25,000+ datasets across 70 fragmented portals into a single discoverable catalogue.
Sep 2024 · 6 minRead →Practical sessions on AI agents, data readiness and sovereign deployment, run with partners including techUK and The Data Lab.
What separates a chatbot from an agent that can actually complete multi-step work, and what your data estate needs to look like before agents can be trusted with it.
In partnership with techUK
Publishing data openly was built for human readers. AI agents need provenance, quality signals and machine-readable context before they can use it. This session looks at what has to change.
With The Data Lab
A practical walk through how deep agents plan, use tools and recover from failure, and where reliable metadata makes the difference between a demo and a dependable system.
With The Data Lab
AI-readiness across industries. Two 30-minute conversations on LinkedIn Live, with audience questions and recordings available afterwards.
Dr Hemant Tripathi, Founder, Syntropic
Data gaps, data justice and AI-readiness in nature and climate.
Younghwa McLean, Senior GRC Leader
Data trust, permissions and accountability: what must be in place before AI agents act, and how existing governance frameworks address the EU AI Act.
Plain definitions of the terms that come up in every sovereignty and AI-readiness conversation. Free to cite and link.
Data sovereignty means an organisation's data never leaves its own environment. AI sovereignty extends the same principle to inference: the AI model runs inside that boundary too, rather than sending data to a third-party commercial API. Together they allow an organisation to adopt AI without relinquishing custody of its data, which is why it matters most in government, healthcare, defence and financial services.
AI-native enrichment is the use of AI models to automatically generate descriptive metadata, score data quality, detect sensitive information and restructure raw files, rather than performing these tasks manually or with rule-based scripts. The distinction from "AI-enabled" is architectural: an AI-native tool is built around model inference as the primary mechanism, not as a feature added to an existing catalogue.
Agent-ready data is data an autonomous AI agent can find, interpret and use without human intervention. It requires machine-readable metadata, explicit provenance, quality signals and clear usage rights. Publishing data openly is not sufficient: without this context an agent cannot judge whether a dataset is relevant, current or trustworthy, so it either ignores it or uses it badly.
DCAT3 is the Data Catalog Vocabulary, version 3, a W3C standard for describing datasets and data services so they can be shared between catalogues and organisations. It is the basis of the UK Government Metadata Exchange Model and is widely used across European public-sector data infrastructure, which makes it the practical default for cross-government data sharing.
Croissant is an open metadata format from MLCommons that describes machine learning datasets, combining metadata, resource descriptions, data structure and default ML semantics in a single file. Built as an extension of Schema.org, it makes datasets portable across ML frameworks such as PyTorch, TensorFlow and JAX, and discoverable beyond the repository hosting them.
Dark data is data an organisation collects and stores but never uses, typically because it lacks the metadata needed to find, understand or trust it. Industry estimates from Gartner, IBM and Splunk place dark data at between 55% and 80% of data holdings, representing both a wasted asset and an ongoing storage, governance and compliance cost.
Reading about metadata enrichment only goes so far. A free 30-minute demo shows it running on real data, and takes no preparation from your side.
Book a free demo