Enterprise Knowledge Architecture Beyond Basic RAG
Embedding documents and retrieving nearby chunks is useful, but enterprise questions often need entity identity, relationships, business vocabulary, effective dates, permissions and the authoritative source behind every answer. This six-part series moves from retrieval to knowledge architecture: when a knowledge graph earns its keep, multimodal retrieval across tables, images and diagrams, ontologies without the academic fog, data lineage for AI answers, and correcting a knowledge base without carrying the error forward. It builds on CloudNala's RAG That Survives Production series rather than repeating it. It follows Blouberg Infrastructure Programme, a fictional public infrastructure programme with contracts, reports, maps, site photographs and invoices; its data and figures are illustrative. Research claims were checked on 5 November 2026.
6 articles
- Part 1RAG retrieves similar text. Enterprises need trusted meaning.A retrieval pilot can find passages that look right and still give a wrong answer, because enterprise questions depend on identity, relationships, definitions, dates and authoritative sources. Part 1 of Enterprise Knowledge Architecture Beyond Basic RAG sets out what a trusted answer needs beyond similarity.9 min read
- Part 2When you need a knowledge graphA relationship-heavy question does not automatically need a graph database. Part 2 of Enterprise Knowledge Architecture Beyond Basic RAG sets out a five-rung ladder from a simple lookup to a shared RDF knowledge graph, the signals that justify each rung, and why identity is the real cost.8 min read
- Part 3Multimodal RAG is an information-architecture problemTables, site photographs, engineering drawings and maps each carry their truth in a different place, and a model that can read images does not change that. Part 3 of Enterprise Knowledge Architecture Beyond Basic RAG shows how to route each kind of content to the representation it needs.8 min read
- Part 4Ontologies without the academic fogAn ontology is a written, machine-readable agreement about what things are and how they relate. Part 4 of Enterprise Knowledge Architecture Beyond Basic RAG shows how to build one from real questions, in levels a delivery team can adopt one at a time, with owners and states instead of philosophy.8 min read
- Part 5Data lineage for AI answersLogging what a retrieval step returned is not enough to explain an answer whose evidence passed through scanning, extraction and a graph first. Part 5 of Enterprise Knowledge Architecture Beyond Basic RAG shows how to record answer provenance in both directions, using the W3C PROV model, at a depth that matches the risk.8 min read
- Part 6Correcting the knowledge base without preserving the errorWhen a fact in an AI knowledge base turns out to be wrong, it has usually been copied into chunks, graph edges, summaries, caches, test sets and answers already given. Part 6 of Enterprise Knowledge Architecture Beyond Basic RAG sets out how to correct at the origin, keep the record and reach everyone affected.7 min read
Working through any of this in your own organisation?
These articles come out of delivery work rather than theory. If the problems in this series look familiar, a short scoping conversation is usually the quickest way to work out what is worth doing first.
Talk to CloudNala