Proposal: free ontology engineering course for Learn > Courses

Proposed entry, Learn > Courses

- [Ontology Training: Taxonomies, Knowledge Graphs and AI](https://tesseract.academy/courses/ontology-training-knowledge-graphs-complete-course/): A free 45-lesson course on RDF, OWL 2, SPARQL, SHACL and knowledge graph construction, including how to constrain LLM output against a published vocabulary.

I am the author and I run the organisation that publishes it. Opening an issue rather than a PR, per CONTRIBUTING.

Pillars. 1 and 2. CURATION.md puts "high-quality books, courses, and tutorials that teach developers these subjects" in scope, and lists evaluating and operating AI applications under pillar 2.

The gap it fills. None of the ten current Courses entries covers knowledge representation: no RDF, no OWL, no SHACL, no knowledge graphs. That absence is checkable in your own README. It matters directly for two sections you already maintain. Retrieval and data lists LlamaIndex and Haystack, which solve retrieval; Evals and reliability lists evaluation tooling. Neither addresses whether the entities a pipeline extracted actually exist, which is the failure this course teaches developers to close.

The developer problem. Retrieval hands a model relevant text. Nothing in that loop can tell the model the entity it just asserted does not exist. An ontology plus a closed-world check can, at the term level, without a judge model.

Primary sources, all reproducible:

• The SHACL gap, measured across three real vocabularies: open-world SHACL validated as conformant all 300 data graphs seeded with a fabricated term, drawn from 418 fakes. A closed-world vocabulary gate caught all 300 with no false positives on clean data. https://gov.tesseract.academy/research/ontology-correctness-benchmark
• Constraining a model against a published vocabulary: fine-tuning Qwen3-Coder-30B on IES4, a UK government information exchange standard, cut confabulated ontology terms from 93.7% to 1.0% and raised term conformance from 0% to 88.6%, verified against the ontology rather than by an LLM judge. https://gov.tesseract.academy/research/fine-tuning-llm-government-data-standard
• The engine used in the course is open source: https://github.com/fabio-rovai/open-ontologies
• The syllabus above is machine-readable RDF, SHACL-validated in CI, so the coverage claim is verifiable by query: https://github.com/fabio-rovai/ontology-curriculum

Track record. We built the AI ontology extension generator for the UK National Digital Twin Programme, published under Apache 2.0 in the National-Digital-Twin GitHub organisation. The course has around 100 registrations.

Maintenance. Current, free, no payment step and no certificate upsell.