Documentation
Knowledge Graph
A Knowledge Graph is a connected model of your business: the entities you operate on, such as customers, locations, invoices, and jobs, plus the metrics, rules, and relationships that link them. SIGNLD builds it from read-only connections to your systems, so questions are answered against your structure rather than a generic guess.
How it works
- Connections read schema and rows from each source system. Nothing is written back.
- Entity resolution matches the same real-world thing across systems, so one customer is one entity even when three systems spell the name differently.
- Relationships record how entities connect, for example invoice to customer to location to owner.
- Business rules and metric definitions are attached to entities, so a term like active customer means one thing across every answer.
- The graph is temporal. It records how definitions and values changed, which is what makes month-over-month comparisons honest.
How to use it
- Review proposed entities early. Correcting one bad match saves a dozen confusing answers later.
- Add the systems that disagree with each other. Conflict is where the graph earns its keep.
- Keep definitions in the graph rather than in the question, so every answer uses the same logic.
Building the graph is not a prerequisite project. It starts on the first connection and improves as you confirm proposals. See onboarding.
Knowledge Graph FAQ
How is a Knowledge Graph different from a data warehouse?
A warehouse stores rows for querying. A Knowledge Graph stores what those rows mean: entities, relationships, metric definitions, and business rules. SIGNLD uses the graph so an AI answer can be computed against your definitions instead of inferred from column names.
Do I need a data team to build it?
No. SIGNLD proposes entities, metrics, and relationships from your connected sources, and a business owner confirms or corrects them. No modelling project is required up front.
What happens when two systems define a metric differently?
The conflict is surfaced rather than silently resolved. You choose the definition of record, and every future answer uses it, with lineage showing which rows contributed.
Further reading
Was this helpful?
Contact support