Vinoo Ganesh discusses why financial users need to trace AI-generated facts and calculations to their sources. He separates errors from retrieving the wrong document, extracting the wrong figure and relying on a model to perform arithmetic.
He describes Kepler's approach of preprocessing facts into structured references, using the model to plan retrieval and using code for lookups and calculations. A subsequent check verifies, corrects or refuses outputs that cannot be tied to sourcesEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria..
Ganesh also discusses broad retrieval followed by rerankingRetrieval is the process of selecting relevant stored information and returning it to an AI system for the current task., tenant-scoped document accessAn agent permission boundary defines and enforces which resources and actions an automated agent is allowed to use. and questions builders can ask about citations, arithmetic and unsupported facts. His examples and performance claims reflect his own company's system.
Watch on YouTube
