Mortgage Intelligence Platform
Understand mortgage at its full depth.
Expert mortgage understanding across documents, data, policy, calculations, exceptions, and evidence, across the mortgage lifecycle. Mortgage Quality Control is where it starts.
Built to show what the system read, what applied, what it calculated, what remains unresolved, and the evidence behind the result.
- SOURCEPaystub, 03/2026page 1 of 2, position kept
- FACT / CONTEXT$8,940 / mo qualifying incomeYTD annualized · borrower 1 · four sources
- POLICYReconciliation rule v2.3.0selected by program, active on the run date
- CALCULATIONSpread 3.7%, limit 5%W-2 $8,875 · 1003 and 1008 $9,200
- DECISIONRaised, then acceptednot a defect; reviewer accepted the basis
- EVIDENCEKept with the findingvalues, pages, rule version, disposition, date
The category
Documents contain facts. Mortgage Intelligence understands what they mean.
A mortgage is not one document and not one field. The meaning of a value depends on the borrower, the property, the program, the timing, the policy, the calculations, the other documents in the file, and what has already happened in the lifecycle. Mortgage Intelligence is the layer that connects those things.
Document intelligence
Questions about a page.
- What document is this?
- Where is the value?
- What text is on the page?
- What field was extracted?
Mortgage Intelligence
Questions about the mortgage.
- Which value actually governs?
- Which policy applies on the relevant date?
- Should two sources agree?
- What is missing?
- Is the difference legitimate or a defect?
- Which calculation should run?
- What action follows?
- What evidence supports the conclusion?
Reading is necessary. Understanding is the differentiation.
The lifecycle
One intelligence layer. Across the mortgage lifecycle.
The same mortgage is examined repeatedly by different teams for different reasons. The application changes. The need for coherent, evidence-backed mortgage understanding does not.
- Origination
- Underwriting
- Closing
- Quality controlFirst application
- Delivery
- Servicing
One shared intelligence substrate
- Documents
- Mortgage data
- Policy
- Rules
- Calculations
- Evidence
- History
- Exceptions
Not every stage has an application today. Mortgage Quality Control is the first, and the only one available for evaluation. The others are stated below with their actual status.
The applications
One intelligence layer. Purpose-built applications.
Five applications share the same platform and address different decisions across the lifecycle. Each card carries its status, and only one is available to evaluate today.
Mortgage QC
Quality controlWhole-file quality control with cross-document synthesis, versioned controls, evidence-linked findings, and reviewer disposition.
Explore Mortgage QC →PlannedUnderwriting
UnderwritingMortgage reasoning for income, assets, liabilities, credit, collateral, eligibility, conditions, policy, and exceptions.
See the direction →PlannedClosing
ClosingIntelligence for disclosures, settlement readiness, closing conditions, document consistency, timing, and execution requirements.
See the direction →PlannedDelivery
DeliveryIntelligence for investor readiness, mortgage data alignment, delivery requirements, package completeness, and evidence-backed handoff.
See the direction →ResearchServicing
ServicingMortgage intelligence carried forward into servicing policy, lifecycle events, exceptions, compliance, and defensible downstream decisions.
See the direction →Statuses are set by product, not by marketing. What each status means.
First application
Mortgage Quality Control is where the platform proves itself first.
QC forces intelligence to be accountable. A useful finding must state what happened, which requirement applied, what evidence supports it, what could not be established, and what a reviewer did next.
Intake
Loan file received and queued.
Classification
Documents identified and segmented.
Extraction
Values located, kept with page and source.
Synthesis
Reconciled across the file; conflicts recorded.
Validation
Versioned rules run. Skips recorded too.
Review
Exception raised. A reviewer resolves it.
Evidence
Finding kept with the rule version that made it.
Look at the third number. Those rules could not run, because a field they needed was missing. They are reported separately from passes, because not checked and checked and fine are different answers, and a reviewer deciding where to spend an hour needs to tell them apart.
How the intelligence works
Use models for judgment where judgment belongs. Use deterministic systems where truth can be computed.
Seven components underneath every application. MIM, the Mortgage Intelligence Model, helps interpret mortgage information in context. GMKL, the Governed Mortgage Knowledge Layer, establishes what applies. Deterministic tools calculate and enforce. Authorized people resolve high-impact judgment. The evidence record connects the result back to each of them, and MIB, the Mortgage Intelligence Benchmark, measures the whole on mortgage work.
Governed mortgage data and standards
A normalized, source-linked representation of mortgage facts and lifecycle context, aligned to industry data standards.
Governed Mortgage Knowledge Layer (GMKL)
Effective-dated policy, program requirements, rules, exceptions and overlays.
Mortgage Intelligence Model (MIM)
The intelligence layer that helps interpret mortgage data, requirements, evidence and decisions in context, across documents, policy and the lifecycle.
Rules, calculations and decision intelligence
Deterministic calculations and validation where correctness should not depend on generated prose.
Evidence and explainability
Traceability from material conclusions back to source data, documents, rules, calculations and reviewer actions.
Workflow and remediation
Where a named person reviews, corrects and disposes of each finding, and the decision is recorded.
Mortgage Intelligence Benchmark (MIB)
Mortgage-specific testing of correctness, applicability, evidence and professional work products.
Trust
Intelligence you can examine.
A system that reaches a conclusion a lender must defend has to show its working, and a lender's obligations now extend to the AI its vendors use. Fannie Mae's LL-2026-04 and Freddie Mac's Bulletin 2025-16 both require a documented AI/ML governance program and reserve the right to ask what safeguards are in place.
Source-aware
Every material fact retains its source and context.
Policy-aware
Applicable requirements are versioned and effective-dated.
Calculation-aware
Deterministic mortgage math runs as code when it can be computed.
Evidence-backed
Findings retain the evidence and the rule or tool record behind them.
Human-governed
Unresolved judgment and exceptions are escalated to a named person, not silently invented.
Data-controlled
Customer documents are not used as general training data without an explicit written agreement.
The evidence record exists as a by-product of the review, not as a document written afterwards. Nothing here implies approval, certification or endorsement by Fannie Mae, Freddie Mac or any agency. Security and data handling, including what is current and what is still being built.
The platform in action
Why did the system reach this conclusion?
One value, followed from the page it was read on to the reviewer who disposed of it. The figures are from a synthetic file and are illustrative; the chain is the one the product keeps.
- 01 · Source
Paystub, March 2026, page 1 of 2
Read from the page, and stored with the document, the page and its position on the page.
- 02 · Fact / Context
Gross monthly income $8,940, placed in the file: qualifying income, Borrower 1, four sources
Year-to-date annualized, confidence retained. Paystub $8,940. W-2 $8,875. The 1003 states $9,200 and the 1008 used $9,200 to qualify. The disagreement is carried forward, not resolved silently.
- 03 · Policy
Income reconciliation rule, version 2.3.0
Selected by loan program: conventional 30-year fixed. The version that was active on the run date is bound to the result.
- 04 · Calculation
Spread 3.7% against a 5% tolerance
Deterministic code with the formula and inputs shown. The same inputs and version always produce the same result.
- 05 · Decision
Raised for a reviewer, then accepted: not a defect
Inside tolerance, but the 1008 used the stated figure rather than a derived one, so the basis is a reviewer’s call. A named reviewer accepted it on 2026-09-08: the underwriter used the stated figure with a written calculation on file.
- 06 · Evidence
Everything above, kept with the finding
Four values with their pages, the rule and its version, the calculation, the disposition and its date, as a by-product of the review rather than a document written afterwards.
Evaluation
We measure the platform on mortgage work.
Generic AI benchmarks do not tell a lender whether a system can reconcile income, identify the applicable requirement, determine that information is missing, explain an exception, or produce evidence a reviewer can defend. MIB, the Mortgage Intelligence Benchmark, is built around representative mortgage cases and professional work products.
We do not publish performance numbers without the method, the document set and a comparison anyone can reproduce. Until then, the honest test is your own files with outcomes you already know. Explore our research approach.
Put the Mortgage Intelligence Platform against a real file.
Start with a synthetic, sample, or appropriately controlled historical file. See what the system read, which controls ran, what could not be established, what evidence remains behind a finding, and where human judgment enters.