Automated Underwriting System (AUS)

Non QM eligibility depends on connected information across the borrower, property, credit, income, liabilities, program rules and supporting evidence.

A loan may meet the LTV requirement but fall below the minimum DSCR. A value entered by the broker may conflict with a verified document. Evidence may be missing, expired or rejected.

The product needed to answer four questions clearly:

1. What does the system know about this loan?

2. Is it eligible for the selected program?

3. What evidence is still required?

4. Why did the system reach this decision?

The challenge was turning underwriting logic into a sequence brokers could understand and act on without asking them to interpret a rulebook.

I organised the flow around how certainty develops during pre underwriting:

Application overview → Eligibility summary → Initial submission requirements → Findings report

The application establishes the facts. Eligibility evaluates them against the selected program. Submission gathers the evidence needed to support the decision. Findings preserve the reasoning in a form that can be reviewed and shared.

Each stage has a different job, but all four carry the same loan context forward.

The Application Overview brings borrower information, loan and property details, assets, liabilities, credit data and the selected program into one reviewable space.

Clear information groups help brokers verify the scenario without returning to the original application. A persistent summary keeps essential details such as program, loan amount, LTV, DSCR and borrower visible throughout the workflow.

This stage is deliberately factual. It confirms what the system will use before those inputs become an eligibility decision.

Once eligibility is understood, the product needs evidence to support it.

Brokers may receive a complete package as one PDF or collect documents individually. I designed both paths to feed into the same validation model. The system analyses the upload, identifies the documents and maps them to the scenario’s requirements.

The experience then shifts from file upload to submission readiness. Each requirement is marked as received, pending, expired, rejected or duplicated. Progress and summary counts expose what remains unresolved.

Exceptions include a recovery path. Missing evidence can be uploaded, rejected documents replaced and duplicates reviewed. The loan can be registered only when the minimum requirements are met, making readiness a visible product state rather than a judgement the broker makes alone.

The work created a connected pre underwriting experience where each stage increased the quality of the decision.

Loan information became reviewable context. Eligibility became explainable rather than binary. Documents became evidence rather than attachments. Findings became a clear account of what passed, what required attention and why.

The structure can be expressed simply:

Facts establish the scenario. Rules evaluate it. Evidence supports it. Findings explain it.

This supports Prudent AI’s broader promise of moving underwriting clarity earlier in the Non QM workflow without presenting company level outcomes as metrics from this specific design.

Trust in an AI product does not come from presenting a confident answer. It comes from making the path to that answer visible.

For Upfront AUS, that meant preserving context, explaining how requirements were evaluated and giving every exception a consequence and a next action.

Automation feels reliable when people can see the facts, reasoning and recovery behind the result.

If I continued the work, I would test the flow across a wider range of incomplete and contradictory loan files, then measure whether it reduces avoidable resubmissions and manual clarification before underwriting.