Librarian — The Knowledge Curator

The Librarian agent is ikigize's AI specialist for knowledge mapping and resource discovery. It operates across two deeply connected roles: connecting platform entities to the Learning Graph so that recommendations, search, and discovery work automatically — and actively finding high-quality public resources that match those mapped topics.

Two Roles, One Agent

What the Librarian Does

Connecting the knowledge graph and enriching the resource library

Entity Cataloging

Maps any platform entity to Topics and Skills in the Learning Graph

Runs a streaming search-explore-commit pipeline

Finds existing matching nodes before proposing new ones

Unlocks library recommendations and graph-augmented search instantly

Relationships labelled with 'catalog' source for transparency

Resource Discovery

Searches the web and curated sources for public learning resources

Context-aware: uses entity topic mappings and user language preferences

Evaluates credibility, relevance, and difficulty level

Saves discovered resources directly to the library with full metadata

Each saved resource also enriches the Learning Graph

Quality Curation

Evaluates source credibility and content quality

Matches difficulty to the learner's current context

Selects across formats: video, articles, tools, documentation

Provides comparative analysis of multiple sources

Smart Categorization

Difficulty level classification (beginner / intermediate / advanced)

Content type identification (read / watch / listen / use)

Topical tagging aligned with the Learning Graph

Auto-links resources to topics via DEVELOPS on save

Entity Cataloging: Connecting to the Learning Graph

The most foundational capability of the Librarian is entity cataloging — mapping a course, campus, module, organisation, or resource to the Topics and Skills it relates to in the Learning Graph. Once mapped, the entity immediately benefits from:

Graph-augmented library recommendations via multi-path topic traversal
Graph filter panel allowing users to filter its library by topic
Cross-entity discovery — the entity appears in searches for its mapped topics
Profile matching — learners interested in those topics can be connected

The Cataloging Pipeline

The Librarian performs full-text search across the existing Learning Graph for Topics and Skills that match the entity's name, description, and metadata.

•Composite full-text and keyword search across Neo4j
•Locale-aware field boosting for best match quality
•Covers both exact matches and semantic near-matches

Explore

For each promising candidate, the Librarian explores the surrounding subgraph — looking at parent topics, related concepts, and sibling nodes to find the most precise and complete set of mappings.

•Traverses SUBTOPIC_OF and RELATED_ESSENTIAL_TOPIC relationships
•Retrieves graph context for LLM reasoning
•Identifies the right level of specificity (not too broad, not too narrow)

Deduplicate

Before proposing any new Topic nodes, the Librarian checks whether a matching node already exists. This prevents ontology fragmentation and keeps the graph clean.

•Exact name matching across all locales
•Similarity scoring against existing nodes
•New nodes only proposed when genuinely novel

Commit

Verified Topic mappings are written to the graph as DEVELOPS, FOCUSES_ON, or INTERESTED_IN relationships — depending on the entity type. New Topic nodes are created if needed.

•Relationships labelled with RelationshipSource: 'catalog'
•Educators can review and adjust all committed mappings
•Immediately activates all downstream library features

Streaming Transparency

The cataloging pipeline runs as a real-time streaming process (Server-Sent Events). Users see a live progress bar moving through each stage — search, explore, deduplicate, commit — making the AI process visible and understandable rather than a black box that either works or doesn't.

Relationship Source Labels

Every entity-to-graph relationship created by the Librarian carries a source label:

Source	Set by	Trust level
`manual`	Educator or admin directly	Highest — always honoured
`catalog`	Librarian agent pipeline	High — reviewed on demand
`resource_save`	Inferred when a resource is saved	Automatic enrichment

Editors can see how each mapping was created and can override or remove Librarian-generated mappings at any time.

Resource Discovery

Once an entity has topic mappings, the Librarian can search for public resources that DEVELOP those exact topics. This is how the platform's global resource pool grows.

How It Works

The Librarian opens with the entity's current topic mappings as context. You can refine the focus with a free-text prompt — for example: "find beginner-friendly video tutorials on transformer models". The Librarian then:

Uses the entity's mapped Topics to anchor its search
Searches the web, YouTube, and curated resource databases
Evaluates each result for quality, relevance, difficulty, and credibility
Saves the top resources directly to the library with full metadata
Automatically maps each saved resource to the relevant Topics via DEVELOPS

Double Enrichment

Every resource the Librarian discovers enriches the library AND the Learning Graph simultaneously. The resource's topic mappings become part of the permanent graph — meaning future graph traversals and recommendations will include it, benefiting every entity that shares those topics.

Resource Types Covered

Video (Watch) — tutorial channels, official product demos, conference talks, educational content creators. Evaluated for production quality, clarity, and community engagement.

Written (Read) — articles, official documentation, tutorials, educational platform content. Prioritised by authority, recency, and clarity of explanation.

Tools (Use) — widely-adopted, well-maintained software and frameworks. Checked for active community support, documentation quality, and clear use cases.

Audio (Listen) — podcasts and lectures where appropriate to the subject matter.

Context-Aware Discovery

The Librarian adapts its search strategy to the current context:

Task-level — highly targeted, practical, immediately actionable resources
Module-level — broader and more comprehensive; foundational + advanced coverage
Campus/Org-level — domain-spanning resources that serve the full community
User language preference — results prioritised in the user's preferred language

Quality Evaluation Criteria

Relevance

Resources must directly address the mapped Topics at an appropriate depth — not too superficial, not unnecessarily complex.

Source Credibility

Official documentation and authoritative sources prioritised
Reputable educational platforms and recognised experts
Active maintenance and community validation

Difficulty Matching

Level	Criteria
Beginner	Assumes little prior knowledge, gentle curve, lots of examples
Intermediate	Builds on foundations, moderate complexity, some prerequisites assumed
Advanced	Expert-level, high complexity, significant prior knowledge assumed

Adaptive Selection

The Librarian reads context clues — module complexity, task descriptions, entity scope — to match difficulty automatically. Educators can always override the classification.

The Growing Global Library

The Librarian's work extends well beyond individual discoveries. Everything it finds contributes to a shared, growing knowledge base:

Compounding value — resources discovered for one course become available to every other entity that maps to the same topics. No redundant searching.

Graph enrichment — every resource saved enriches the graph topology. New DEVELOPS links feed into traversal algorithms, improving recommendations for everyone.

Community quality — as the resource pool grows, the recommendation engine has more signals to work with. The more entities that are cataloged, the more precisely it can match resources to context.

Collective intelligence — the platform learns from resource usage patterns, surfacing the most-engaged resources higher in future recommendations.

Building Together

Every entity cataloged, every resource discovered, and every topic mapping created by the Librarian contributes to a shared knowledge infrastructure. The platform grows smarter with every interaction — not just for you, but for everyone.

Using the Librarian in the Library

The Librarian is accessible from any library context via the Librarian button in the library header. It opens pre-loaded with the current entity's topic mappings and language settings.

Catalog first — if the entity has no topic mappings yet, the Librarian will offer to catalog it before discovery. This one-time step unlocks all downstream graph features permanently.

Provide a prompt — add specific guidance about what type of resources you need. The more specific, the better the results.

Empty state shortcut — when a library search returns no results, the Librarian is surfaced as the primary recovery action, pre-loaded with your search query.

Your Next Steps

Explore how the Librarian fits into the broader platform:

Learning Graph — the knowledge ontology the Librarian builds and uses
Library System — the broader library infrastructure, graph integration, and search layers
Professor Agent — how instructional design and resource curation complement each other
Iki (Dean) — how goal-setting informs what topics and resources are relevant
Agent Overview — understanding how all agents work together