AI Cataloging

The Librarian agent can automatically map any entity to the graph. When triggered, it analyses the entity’s name, description, and metadata and runs a structured pipeline to find matching Topics (and Skills) in the graph. Committed relationships use the catalog source and a fixed strength (e.g. 0.7).

The Cataloging Pipeline

The pipeline runs as a real-time streaming process so users see progress instead of a single blocking step:

  1. Search — Full-text search across existing Topics and Skills for direct matches. The Librarian uses the entity’s text to build a query and retrieves candidates from the graph.

  2. Explore — Semantic exploration of related nodes to find broader or adjacent concepts. This catches topics that don’t match the exact words but are relevant (e.g. “neural nets” → “Deep Learning”, “Attention mechanisms”).

  3. Deduplicate — Proposed new nodes are checked against existing ones. If the Librarian would create a node that is effectively the same as an existing Topic or Skill, the system links to the existing node instead of fragmenting the ontology.

  4. Commit — Verified relationships are written to the graph with source catalog. The entity (course, module, resource, campus, etc.) is linked to the chosen Topics/Skills with the catalog strength. Users can review, add, or remove these mappings in the mapping editor.

What Gets Mapped

The Librarian can be run on:

  • Courses — infers DEVELOPS topics from title, description, and outline
  • Modules — same for module-level content
  • Sessions, Tasks — when they are exposed to the graph
  • Resources — from title, description, and type
  • Campuses, Organisations — infers FOCUSES_ON from name and description

Each mapping is stored as an entity–topic (or entity–skill) edge with source catalog. Manual edits in the mapping editor create or update edges with source manual, which take precedence in weighting and display.

After Cataloging

Once committed, the entity immediately benefits from:

  • Library recommendations that include it when users search or browse by those topics
  • Cross-campus and cross-organisation discovery
  • Learner matching and “My Map” interest surfaces
  • Structural and member-activity propagation if containment and activity hooks are enabled (see Signals & Sources)

Educators can refine the result by adding or removing topics manually; those changes are preserved and reflected in strength and source.

Next Steps