01
Retrieval without resolution
The right words appear in the search results, but the system cannot tell which document actually governs, which has been superseded, or which of two conflicting policies applies in this jurisdiction.
Semantic Intelligence
Closing the business-data semantic gap.
Semantic intelligence is the capability to resolve a business question against the organisation's explicit model of itself — rather than against a flat corpus of documents that happen to contain the words.
What is enterprise semantic intelligence at Cyberwisdom?
Enterprise semantic intelligence at Cyberwisdom Group is the practice of resolving enterprise questions against an explicit, governed model of the business rather than against unstructured text alone. It is led by Origin Big Data, a group company whose stated purpose is to bridge the business-data semantic gap and build enterprise semantic intelligence infrastructure. The capability combines business ontology (HAKDB and R-LED-R), managed GraphRAG through ContextOrgin, ontology-based knowledge management through Kora KM, and governed delivery through VibeChat. The result is that an AI system's answer can be traced to a source, a rule and a version, and that the organisation can tell the difference between a fact, an inference and an absence of evidence.
The gap
Three ways enterprises lose their grip on meaning.
These are the specific failure modes that separate an AI system that can be trusted from one that cannot.
02
Structure without semantics
The data is beautifully modelled and correctly normalised — and still cannot answer a business question, because meaning was never modelled alongside it.
03
Freshness without signal
Every source is technically available. The problem is that nothing marks which of them was true last week and which was changed this morning, so a stale answer is delivered with complete confidence.
The capability
What semantic intelligence actually adds.
Resolution over similarity
The system resolves entities, rules and relationships in the ontology, so it can reason over the structure of the business rather than the shape of a document.
Authority by declaration
Which source governs which question is declared in the model, not inferred from ranking or recency.
Currency as a first-class attribute
Freshness is tracked and enforced, so a superseded policy cannot reach a customer through a confident answer.
Grounding that is structural
The ontology is wired into retrieval, so grounding is a property of the architecture rather than an instruction in a prompt.
Compositional reasoning
Questions spanning multiple entities — a customer, a policy, an obligation, a jurisdiction — can be answered because the relationships between them are modelled.
Evidence with lineage
The answer traces back through the graph to the underlying sources, with the versions and rules that produced it.
Products
The tooling behind the practice.
ContextOrgin
Enterprise GraphRAG management system — operating and governing the graph retrieval layer that semantic intelligence depends on, including the graph's structure, freshness and access model.
- Managed graph operations
- Retrieval governance
Kora KM
Ontology-based knowledge management system — institutional knowledge modelled against business entities and rules, with ownership, versioning and currency built in.
- Ontology-grounded knowledge
- Ownership & change control
VibeChat
Intelligent AI knowledge copilot — delivering resolved, grounded answers to people at the point of work, with the sources shown.
- Grounded answers with sources
- Knowledge discovery
LyndonAI
The enterprise AI platform where context assembly and ontology management are governed alongside agent deployment and evaluation.
- Context assembly
- Evaluation & evidence
Measuring it
How you know it is working.
Semantic intelligence has metrics, and they should be on the dashboard
Most organisations measure model accuracy. Far fewer measure whether the system was given the right material to be right. Both need to be visible.
- <b>Coverage.</b> What proportion of the business domain is actually modelled? An unmodelled area is an area where the system is guessing.
- <b>Resolution rate.</b> Of the questions asked, how many are answered by resolving the model rather than by falling back to free-text search.
- <b>Authority accuracy.</b> When sources conflict, does the system pick the correct one according to the declared model?
- <b>Freshness compliance.</b> How often is a context assembly made up of sources that were current at the time of use?
- <b>Citation integrity.</b> Do the cited sources actually support the claim made, and would a reviewer agree?
- <b>Grounding impact.</b> What does accuracy look like with full context versus without? That number is the business case for the entire layer.