Mortgage Lending
Explore how the AI Optimizer grounds responses in mortgage application data and lending-policy documents. Work through a small portfolio of labelled mortgagors, such as MORTGAGOR_001, and see how the answers change as structured data and documents are added.
Run this use case in a clean demo environment, such as the one created by the Walkthrough. It imports mortgage prompts, creates and resets mortgage database objects, and adds a vector store for policy and application documents.
The lender, applications, and criteria are demonstration data only: they contain no PII and do not produce lending decisions.
What You'll Explore
| Step | Settings | What it demonstrates |
|---|---|---|
| LLM-only | NL2SQL off, Vector Search off | The model has no knowledge of this lender or portfolio. |
| NL2SQL | NL2SQL on | Live portfolio questions, including average salary by location, LTV, LTI, and expenditure. |
| Vector Search | Vector Search on, NL2SQL off | Lending criteria, evidence standards, and application-pack context. |
| Both together | NL2SQL on, Vector Search on | A screening discussion that compares a mortgagor's data record with policy documents. |
The policy covers income evidence, expenditure, loan-to-value, loan-to-income, and affordability resilience; the kinds of factors considered in responsible lending.
After exploring the grounded workflow, extend it with Deep Data Security to see row- and column-level access enforced on NL2SQL queries.
Complete the Setup before you begin.