Concepts
The words I use most, in the sense I use them. Each one links to the texts where it does work.
- Data model
The minimum shared truth an organization needs to represent consistently in order to understand itself.
Far more than table schemas or entity-relationship diagrams. Joe Reis puts it well: a structured representation that organizes and standardizes data to guide human and machine behavior, inform decisions and enable action.
Before any table, I ask the five Ws: who the model serves, why it needs to exist, what entities and relations it holds, when it covers, where the data lives, and how it will be built and governed. Jumping straight to the logical level is design, not modeling.
Read moreData engineering is building trapsFundamentals outlast wavesModeling is not architectureOriented to data, not only by dataA machine that doesn't know how to remember
- The ladder of expressiveness
Tags, controlled vocabulary, glossary, taxonomy, thesaurus, ontology: each step says more about what data means.
Semantics can be expressed at many levels. At the bottom, loose tags. Then a controlled vocabulary, a glossary of terms with definitions, a taxonomy that orders them, a thesaurus that relates them, and finally an ontology that makes the relations formal enough to reason with.
Each rung costs more to build and keep alive. Climb only as high as the use demands. But remember why you climb: data is about increasing knowledge, not just answering immediate questions.
- Ontology
A formal, explicit specification of how a domain is conceptualized: its concepts, relations, instances and rules.
An ontology defines terms by describing how they relate to other terms, so that machines and people can interpret data the same way across systems.
One lesson I keep from studying the field: the opposite of an ontology isn’t no ontology, it’s a bad one. Every system already embeds one. The difference is whether you notice you’re making it.
Read moreOriented to data, not only by dataThe mystery of knowledge
- Knowledge graph
A graph that is also a representation of knowledge: entities as nodes, relations as edges, and their meaning captured for reasoning.
Born from the semantic web (RDF, Linked Open Data), knowledge graphs carry not just connections but what the connections mean. Combined with vector embeddings, they seem to be the missing piece that lets AI sustain understanding instead of guessing it from raw data.
Read moreOriented to data, not only by dataThe mystery of knowledge
- Semantic layer
Where business concepts, metrics and relations are defined once, so any system, person or agent knows what data means and how to use it.
Semantics should be self-describing and machine-readable. A semantic layer sits between the raw tables and everyone who consumes them, and with AI agents in the loop it stops being a BI convenience and becomes the organization’s shared vocabulary.
Read moreOriented to data, not only by dataThe mystery of knowledge
- Data product
Data built for a purpose and exposed as valuable in itself, with an owner responsible for making it usable.
Data can flow through systems that make things from it. But it can also be the thing itself: modeled, built for purpose and published. Data products move the burden of cleaning and linking from whoever consumes data to whoever produces it, what Tony Seale calls inverting the cost of data integration.
Read moreThere is no generic data
- Data contract
An agreement on what the data is and how its quality is enforced, made where the data is produced.
Three parties share the data value chain: producers who generate it, platforms who keep the infrastructure, and consumers who use it. A contract enforces schema and meaning at the source instead of patching problems downstream: the idea of shifting left.
Read moreThere is no generic data
- Data mesh
An operating model where business domains own their data and publish it as products, under federated governance and on a shared platform.
Centralization promised one place for all data and ran into the limits of one team holding all the complexity. Data mesh distributes ownership to the domains that understand the data best, and keeps it coherent through shared standards rather than a single gatekeeper. I spoke about it at the second Meetup de Dados Curitiba.