Whitepaper

Knowledge under governance: hybrid retrieval, linked storage and a code-aware knowledge graph

How Camaan's knowledge module keeps enterprise documents under the customer's control: it can index without storing the text, fuses keyword and vector search, and builds a knowledge graph under a customer-written ontology. Includes a retrieval evaluation with a negative result reported as found.

Linked storage: an index of embeddings and term counts with no passage text, fetched from the source at read time and checked against its hash.
Hybrid retrieval evaluated on 1,720 queries, with the finding that keyword search alone beat fusion with the shipped embedder.
A knowledge graph built under a customer-authored ontology, with deterministic code parsing and a human-reviewed entity-merge queue.
On 100 multi-hop questions, adding a graph walk raised all-evidence-in-top-ten from 0.49 to 0.60.