A graph governance layer can enrich an LLM prompt with banking context at call time
Recorded on 4.12Internal only
Say while it plays
A graph isn't a chart, it's data stored as entities and the real relationships between them, so a member connects to an account, an account to a transaction, and a transaction to a merchant.
A Groovy policy resolves which member a request belongs to from a task-scoped token, not anything the agent claims, then calls the Graph Context API to fold merchant history and governed definitions into the prompt before the model ever sees it.
The same ontology and precedent get injected into every prompt at runtime, so an agent can discover context on demand without every fact being hardcoded or blowing the context window.
Don't claim
Don't present this as retrieval-augmented generation or document search: the recording opens by explicitly ruling that out, calling it a governance layer above RAG, not a replacement for it.