Image 1 of 1
Foundations Course
What this is
A six-session, cohort-based introduction to knowledge graphs — taught live, in small groups, by people who build them for a living. You start by drawing a model by hand and finish with a working knowledge graph of your own, queried with SPARQL and wired into an LLM.
This is not a video course you watch alone. Every session is live and interactive, built around a rhythm of short teaching blocks, breakout work in threes, and whole-class review. Nobody needs to arrive with the answers. Cohorts draw people from a wide range of professional backgrounds and the course is deliberately built to hold that mix — curiosity matters considerably more than prior expertise, and no coding experience is required.
What you will learn
Think in graphs — model a domain as things and relationships, and say why the open-world assumption changes what your data means
Separate the three boxes — TBox, CBox and ABox, and know which W3C standard governs each
Scope a graph with competency questions — write questions that are grounded, answerable, testable, and that tell you when you are finished
Build a controlled vocabulary that holds — using MECE and Ranganathan's canons, with a coverage model behind it rather than a guess
Write SKOS by hand — mint your own URIs, build a three-level taxonomy, and publish it as valid Turtle
Find the ontology inside your vocabulary — and declare it with owl:Class, rdfs:subClassOf, domain and range
Query the thing you built — write SPARQL that answers the competency questions you set in session two, which is how you know the model is done
Close the neuro-symbolic loop — use an LLM to extract facts into your ontology, and ground retrieval in the graph rather than in embeddings alone
Who it is for
People who need to understand how knowledge graphs actually work, and who would rather build a small one than read about a large one: data practitioners, librarians and taxonomists, architects, analysts, engineers, and anyone whose organisation has started saying the word "ontology" in meetings.
No prerequisites, and no coding experience required. The course starts from what a concept is. Participants come from a wide range of professional backgrounds and consistently find that the mix is part of what makes it work.
How the course runs
Six live teaching days of six and a half hours — five hours of teaching, a one-hour lunch and two fifteen-minute breaks. Sessions are a week apart, so ideas have time to settle between them.
Before the first session there is Session Zero, a two-hour onboarding: tools installed, access checked, and everything that would otherwise be discovered on the morning of session one dealt with in advance. This matters most for corporate cohorts, where locked-down machines are the usual cause of a lost first morning.
Everything is delivered online through Google Classroom, with a dedicated Discord space for the cohort that stays open between sessions. Breakout groups are typically three people. Sessions are not recorded — they are conversations, and they are better for being live.
There is no required homework. Optional reading is provided and many participants take it up, but the course is self-contained.
The six sessions:
Knowledge graph fundamentals — conceptual modelling, classes and instances, TBox and ABox, open versus closed world, and your first model drawn by hand
Controlled vocabularies — competency questions, the CBox, the four kinds of controlled vocabulary, MECE, and building a coverage model for your domain
SKOS and taxonomy — URIs and namespaces, a three-level taxonomy of your own, written out in Turtle — at which point you discover you have been writing RDF triples
Ontology foundations — reading competency questions as structural requirements, ontology design heuristics, and declaring your ontology in RDFS and OWL
Grounding concepts in facts — harmonising vocabulary and ontology, writing your instance data, and querying it with SPARQL
Knowledge graphs and large language models — the neuro-symbolic loop, extraction into your own ontology, entity linking, and GraphRAG
A single domain runs through all six sessions, so by the end the thing you are querying is the thing you started drawing in session one.
What you leave with
A working knowledge graph of your own, built from scratch across six weeks
A controlled vocabulary, an ontology, and instance data — in valid, standards-compliant Turtle
SPARQL queries that answer the competency questions you wrote yourself
A clear sense not just of how things are done but why they are done that way
A cohort of people who built the same thing at the same time, and the Discord space you built it in
What this is
A six-session, cohort-based introduction to knowledge graphs — taught live, in small groups, by people who build them for a living. You start by drawing a model by hand and finish with a working knowledge graph of your own, queried with SPARQL and wired into an LLM.
This is not a video course you watch alone. Every session is live and interactive, built around a rhythm of short teaching blocks, breakout work in threes, and whole-class review. Nobody needs to arrive with the answers. Cohorts draw people from a wide range of professional backgrounds and the course is deliberately built to hold that mix — curiosity matters considerably more than prior expertise, and no coding experience is required.
What you will learn
Think in graphs — model a domain as things and relationships, and say why the open-world assumption changes what your data means
Separate the three boxes — TBox, CBox and ABox, and know which W3C standard governs each
Scope a graph with competency questions — write questions that are grounded, answerable, testable, and that tell you when you are finished
Build a controlled vocabulary that holds — using MECE and Ranganathan's canons, with a coverage model behind it rather than a guess
Write SKOS by hand — mint your own URIs, build a three-level taxonomy, and publish it as valid Turtle
Find the ontology inside your vocabulary — and declare it with owl:Class, rdfs:subClassOf, domain and range
Query the thing you built — write SPARQL that answers the competency questions you set in session two, which is how you know the model is done
Close the neuro-symbolic loop — use an LLM to extract facts into your ontology, and ground retrieval in the graph rather than in embeddings alone
Who it is for
People who need to understand how knowledge graphs actually work, and who would rather build a small one than read about a large one: data practitioners, librarians and taxonomists, architects, analysts, engineers, and anyone whose organisation has started saying the word "ontology" in meetings.
No prerequisites, and no coding experience required. The course starts from what a concept is. Participants come from a wide range of professional backgrounds and consistently find that the mix is part of what makes it work.
How the course runs
Six live teaching days of six and a half hours — five hours of teaching, a one-hour lunch and two fifteen-minute breaks. Sessions are a week apart, so ideas have time to settle between them.
Before the first session there is Session Zero, a two-hour onboarding: tools installed, access checked, and everything that would otherwise be discovered on the morning of session one dealt with in advance. This matters most for corporate cohorts, where locked-down machines are the usual cause of a lost first morning.
Everything is delivered online through Google Classroom, with a dedicated Discord space for the cohort that stays open between sessions. Breakout groups are typically three people. Sessions are not recorded — they are conversations, and they are better for being live.
There is no required homework. Optional reading is provided and many participants take it up, but the course is self-contained.
The six sessions:
Knowledge graph fundamentals — conceptual modelling, classes and instances, TBox and ABox, open versus closed world, and your first model drawn by hand
Controlled vocabularies — competency questions, the CBox, the four kinds of controlled vocabulary, MECE, and building a coverage model for your domain
SKOS and taxonomy — URIs and namespaces, a three-level taxonomy of your own, written out in Turtle — at which point you discover you have been writing RDF triples
Ontology foundations — reading competency questions as structural requirements, ontology design heuristics, and declaring your ontology in RDFS and OWL
Grounding concepts in facts — harmonising vocabulary and ontology, writing your instance data, and querying it with SPARQL
Knowledge graphs and large language models — the neuro-symbolic loop, extraction into your own ontology, entity linking, and GraphRAG
A single domain runs through all six sessions, so by the end the thing you are querying is the thing you started drawing in session one.
What you leave with
A working knowledge graph of your own, built from scratch across six weeks
A controlled vocabulary, an ontology, and instance data — in valid, standards-compliant Turtle
SPARQL queries that answer the competency questions you wrote yourself
A clear sense not just of how things are done but why they are done that way
A cohort of people who built the same thing at the same time, and the Discord space you built it in