Hypothesis
A domain becomes teachable and searchable when it is represented as a connected model — entities, relations, provenance — that updates as the domain moves, instead of a static index.
Problem
Documents pile up; meaning does not. Retrieval without structure returns fragments; structure without provenance returns confident nonsense.
Current state
Active — in production use.
Methods
- Entity and relation extraction into a graph model
- Semantic retrieval over graph neighbourhoods
- Provenance tracking: every node knows where it came from
- Continuous updates as source material changes
Findings
- Graph-shaped knowledge outperforms flat retrieval for curriculum-scale domains
- Provenance is what turns retrieval into something a product can trust