INTELLIGENCE LAYER
Grounded AI over verified data
The Intelligence Layer adds conversational query, grounded narratives, semantic search and labeled predictive models across the product. Its one non-negotiable rule: the data layer supplies every number — the model only writes the prose, and every figure it emits is validated against the source rows before it reaches you.
The assistant is docked in the shell on every signed-in surface — ask it a data question and it runs the query against the verified tables, or press ⌘K to search companies and issuers semantically.
Conversational query
Ask a market or lender question in plain English; the answer renders as a native table over pre-aggregated data, never as an unverifiable claim.
Grounded narratives
Outlier signals, peer comparisons, and market commentary phrased by the model over figures the data layer computed — every number validated before it ships.
Semantic search & scoring
Similar-entity and similar-pool search plus labeled early-warning and prepay models — statistical estimates, clearly labeled, with the model only writing the explanation.
Generated documents
AI-authored dossier decks and reports, where every figure and chart is templated from verified rows and only the connective prose is generated.
THE MODEL SERVING EACH ANSWER IS NAMED ON THE ANSWER AND IS SET BY CONFIGURATION, NEVER BURIED IN CODE. EVERY GENERATED FIGURE IS VALIDATED AGAINST HMDA / GINNIE MAE SOURCE DATA.