databrickslabs / databrickslabs/ontos
[Feature]: LLM generator should also produce glossaries (Concepts), not only ontologies (Classes/Properties)
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 212
- Forks
- 71
- Avg merge
- 4d 10h
- Merged PRs (30d)
- 43
Description
Problem
The Define section offers three creation paths, but they do not cover the full output spectrum evenly:
- Author (manual) — Concepts by default (skos:Concept), can harden to Class/Property.
- Generate (LLM) — produces Classes + Properties only (a structural ontology from table shapes).
- Import (file) — whatever the file contains.
So a user who wants a business glossary (agreed terms + definitions + synonyms, i.e. skos:Concept entries) has no assisted/LLM path — they must author every term by hand. The LLM generator is effectively a power-user-only tool because its output is always formal Classes/Properties.
This surfaced in real use: users find the Class-vs-Concept distinction confusing precisely because the generator hands them Classes when many of them mentally wanted "a vocabulary."
Proposed capability
Give the generator an explicit output-mode choice up front so the user decides what they want to produce:
- Glossary mode — emit
skos:Conceptentries with prefLabel / definition / altLabel (synonyms) / broader-narrower, from the same table/metadata input. Business-vocabulary output. - Ontology mode — current behavior:
owl:Class+rdf:Propertywith domain/range. Formal structural output.
Both land in a concept scheme and both wire to physical assets identically (links key on the IRI regardless of rdf:type), so downstream behavior is unaffected — this is purely about which kind of node the generator mints.
Notes / scope
- Aligns with the glossary -> taxonomy -> ontology maturity ladder the product already frames.
- The generator save path already stamps generated concepts as Draft for review, which applies to both modes.
- Prompt/predicate-palette work: glossary mode constrains the model to SKOS predicates; ontology mode keeps the current owl/rdfs palette.
- Out of scope here: enterprise-scale chunking (separate concern).
Acceptance sketch
- Generator UI has a clear "what do you want to produce?" mode selector (Glossary vs Ontology) before running.
- Glossary mode yields valid SKOS concepts (no owl:Class), with definitions and synonyms populated where derivable.
- Ontology mode unchanged.
- Both save as Draft and appear correctly grouped in Explore.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by tracing the generator UI, its save path, and the prompt/predicate-palette handling described in the issue. Add the Glossary versus Ontology selector, verify glossary output uses valid SKOS concepts with definitions and synonyms where derivable, and confirm ontology output remains unchanged while both modes save as Draft and appear correctly grouped in Explore.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- ai, backend, frontend
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 52/100