What is an ontology?
An ontology explained without jargon, a shared, written-down agreement about what things are and how they relate, and why it is the most valuable asset a data platform holds.
Taxonomy, ontology, knowledge graph
The three terms distinguished clearly, with examples, what each one adds, and which one you are building at any given moment.
Knowledge graphs
What a knowledge graph is, why following relationships answers questions that table joins cannot, and what it feels like to use one.
What is contextualization?
Attaching every piece of operational data to the real-world thing it describes, what that work involves, and what a business gets out of it.
What is a digital twin?
A digital twin explained honestly, a live, connected digital counterpart of a physical operation, why you grow one rather than build one, and how the knowledge graph, time series and events are its anatomy.
The three layers of a model
Assets, functions, and business and operational knowledge, the three kinds of thing a DataHub model describes, and why the top layer is the one that pays.
Naming and standards
How to name resources, labels, external ids and relationships, and the industrial tagging standards that already encode most of your model.
Lineage and data quality
The planned lineage capability, why every derived number will be traceable back to the raw measurements it came from, why that sets how fast an organisation can decide when the world moves, and what it changes about reporting.
What is data governance?
The rules and accountabilities for how an organisation's data is managed, how data sets, policies, functions and the knowledge graph each carry a piece of it, and the four moments where a rule can be enforced rather than merely recorded.
Building your model
A practical recipe for the modelling workshops that produce a DataHub ontology, who to invite, what to decide, and what to leave until later.
Data cleaning
Why bad industrial data does not announce itself, the five defects that actually turn up, how the model around a reading is what convicts it, and the three rules that keep a correction from becoming a worse problem than the error.
Feature extraction
What a feature is, why better features beat better algorithms almost every time, how the event log turns out to be a labelled training set, and why a feature kept as a series compounds while one kept in a notebook does not.
What is machine learning?
Machine learning in plain language for people who will never write any, the methods worth recognising by name, from linear regression through random forests and XGBoost to k-means and neural networks, what each one is actually good for, and how to tell which one a supplier is really selling you.
Synthetic data
Why the failures you most want to predict are the ones you have barely seen, where generated data comes from, the four situations where it earns its place, and the separation rules that keep it from poisoning everything downstream.
Stream processing
Computing on data as it arrives rather than after it has piled up, what a window is and how to choose one, why when a reading happened is not when it turned up, what a computation has to remember, and how a subscription and the SDK do this on the platform today.