KAiM Mortgage Loan Approval Home · Use cases · Mortgage Loan Approval
Status: Live evaluation engine · Demo Data seeded application · No backend; runs in browser
Flagship use case · 03 of 05 · Banking · Mortgage origination

Mortgage loan approval under HELM governance

An AI underwriting assistant has reviewed a single purchase-mortgage application and proposes a high-blast-radius 3-action bundle — deny + send adverse-action notice + close file. This page is a guided demo: pick a persona, review the file, submit the AI's proposal to HELM, see the deterministic decision, then take the next-step action that belongs to your role.

Built to be examined. Every loan decision an examiner — OCC, FDIC, NCUA, CFPB — can read, reason by reason, in the format the exam requires. Accountability, not blind trust.
Life of a loan · where HELM governs — the decision gate is one control point in a workflow that produces examinable records end to end
01
Application
1003 intake · disclosures
02
Processing
docs · credit · income
03
Underwriting decision
approve / deny / counter · you are here
04
Closing
CD · funding · TRID
05
Servicing
boarding · payments

How this walkthrough works

Four steps. Each persona sees the same loan file differently and takes different next-step actions. Try at least two personas to feel the architecture.

1

Pick a persona

Five roles in the same lender. Each has different authority, different views, and different next steps.

2

Review the file

Same application file for everyone — but the framing card shows how your role reads the situation.

3

Submit to HELM

The deterministic gate evaluates 6 checks. The eval is the same regardless of persona — HELM is not opinion-based.

4

Take your action

After the decision, your persona panel on the right shows what you can actually do next. The interactive ones expand inline.

Acting asclick a tile to switch role
1 · Persona 2 · Case 3 · HELM 4 · Action
Pick a persona, then review the case below.
Step 3 · Run the AI's proposal Same proposal, two outcomes. First see what an ungoverned agent does when it simply executes — then run the identical proposal through HELM.
That was HELM. The same agent, ungoverned, would have executed all three actions and reached the borrower — with no underwriter and no record.
HELM· Deterministic Evaluation · 6 checks Evaluating…
What's being evaluated: the AI agent's authority to execute this specific 3-action bundle against this specific application, against fair-lending + adverse-action + documentation + ability-to-repay + sacred-line constraints. Result is the same for every persona — what changes is what each persona can do with the decision.
HELM intervention · System worked as designed

No borrower-facing action executed

4 of 6 gates triggered. No autonomous denial reached the borrower. No regulatory violation incurred. 2 human reviewers assigned. Evidence packet generated. Decision signed and chained.

Per-action disposition

Escalation routing

Evidence packet generated

🔒 Audit Evidence Chain Entry Signed · Immutable
Decision logged. · Switch personas above to see the same decision from another role; their available actions differ.
Why this matters to a regulated lender

Your AI doesn't get the benefit of the doubt. Neither do you.

A mid-size retail bank, a community bank, or a credit union can't answer an examiner with "the model decided." When AI touches a credit decision, the burden of proof is yours — reason by reason, applicant by applicant. HELM is the layer that lets you carry it. Not a policy you promise to follow; a control that is structurally true.

Every decision is reason-coded and reconstructable

No black-box denials. Each decision carries its specific principal reasons, the evidence it rested on, and the config it ran under — signed and chained. An examiner reads it the way you'd want to be read.

Adverse action is compliant by construction

A notice that doesn't meet Reg B §1002.9 — specific reasons, 30-day timing, FCRA score disclosure — cannot send. The gate is the control, not a checklist someone forgot.

Protected-class proxies refuse to compile

ZIP, surname, age, source-of-income type — no tenant can wire a prohibited-basis proxy into decision logic. The sacred line isn't a guideline; the rule engine rejects it. always block.

One record for every regulator

The same signed evidence chain answers OCC, FDIC, NCUA, and CFPB. Model risk (SR 11-7), fair lending (ECOA), and ability-to-repay (Reg Z) all read from one source of truth.

Accountability, not blind trust. Built to be examined.
Built from the
buyer's seat
Why this lens

KAiM's founder spent 25 years inside regulated banking — the Federal Reserve Bank of San Francisco and Wells Fargo's regulatory-remediation program, during the most scrutinized period in the bank's history — building data lineage, model-risk, and continuous-monitoring frameworks. HELM puts governance into the operating model, not on top of it. This is the control we wished existed on the buyer's side of the exam.

This is the wedge. Bring us your loan workflow.

The way KAiM engages is concrete: pick one high-risk AI-assisted decision you're worried about — mortgage underwriting, auto approvals, small-business credit, adverse action, exception handling, deposit/fraud holds. We map it to actor / action / authority / policy / evidence / escalation, then show how HELM gates it. No deck. No demo theater. A working session with your underwriting and compliance teams.