Goal: answers with traceable sources and far fewer hallucinations, by aggregating evidence along explainable paths before the prompt.
The research
Language models answer well only when the right evidence reaches the prompt. We connect text segments, domain concepts and cases in one tripartite graph and assemble the prompt along explainable paths.
What it changes in practice
Answers with sources, in the customer's own documents. The same mechanism carries the decision document: every figure traces back to a process step.