Skip to content

AI Glossary · Concepts

What is Context Graph?

Context Graph

A context graph links entities and relationships with the business definitions and rules an AI application needs to interpret them.

· Updated · Chain of Thought

  • 2 explainers
  • 2 episodes
  • 3 sources

Separate graph structure from business interpretation

On the show, Jitender Aswani uses context graph for a layer that adds business interpretation to entities and relationships. Explaining that distinction, Jitender Aswani says, “Now context graph is building on top of a knowledge graph.” (episode 66, 46:56).

In the same answer he gives the examples that make the term concrete: customer can mean account or opportunity depending on the team, and churn or cohort can carry different definitions across domains and even between people. Graph structure alone does not settle those meanings. The RDF primer describes representing statements as subject-predicate-object triples (W3C, RDF 1.1 Primer), and a business rule is one more statement someone has to model, own and keep current.

One implementation example is Starburst’s AIDA documentation. Its business rules describe how a data product’s metrics are computed, which date column to use and known limitations in the data, and subscribers inherit those rules read-only (Starburst, Business rules). The same page advises fixing unclear column names or wrong join paths at the source instead of patching around them in rules.

Test conflicting meanings even when totals agree

Suppose an agent is asked for “Q2 cancellations” in 2026. Three cancellations exist: A on April 30, B on May 1 and C on July 15, all in UTC. Two calendar rules could apply. Each window includes its start date and excludes its end date.

RuleQ2 windowIncluded events
Calendar yearApril 1 to July 1A, B
Fiscal year starting February 1May 1 to August 1B, C

Both rules return two cancellations, from different events. If you check only the total, you cannot tell which calendar the agent used. Check the included records as well as the number.

For your workflow, tie each question to a named metric definition, calendar, timezone and rule version. When two rules could apply, have the agent ask which one governs the request, or apply a resolution policy that someone has approved in advance.

Preserve origin, ownership and applicable scope

W3C’s PROV overview defines provenance as information about the entities, activities and people involved in producing a piece of data or thing, which others can use to judge its quality, reliability or trustworthiness (W3C, PROV-Overview). For a business rule, that means keeping its source, owner, version, effective dates and the business unit it covers alongside its relationships. Mark which statements come from a source system and which are an approved interpretation.

When a rule changes, rerun the questions it affects and keep a record of which version produced each earlier answer. A graph can hold competing definitions; someone still has to decide which one applies.

Knowledge graph covers the underlying structure, entity resolution covers which records describe the same entity, and data federation covers access across sources. What is context in AI agents? connects the information to what a model sees during a task.

Sources

  • W3C, RDF 1.1 Primer — W3C's introductory primer to RDF, which represents statements as subject-predicate-object triples.
  • Starburst, Business rules — AIDA business rules describe how to interpret a data product's metrics, which date column to use and known data limitations; subscribers inherit them read-only.
  • W3C, PROV-Overview — The 2013 Working Group Note defines provenance as information about the entities, activities and people involved in producing a piece of data or thing, used to assess its quality, reliability or trustworthiness.

Go deeper

From the conversation