luxadoc
MDI · Mortgage Document Intelligence
Every page read,
every number checked.
MDI reads every document in a loan file, checks every figure against the rest of the file and the 1003, and returns an accuracy score it can explain. Inside Encompass, or by API.
- 200+ page packages split and sorted
- 37 document types in the first release
- 1 score per document and per loan
Why it matters
Loan files are checked by hand, and hands miss things.
Reads the whole file
A 200-page package becomes a sorted loan file.
Drop a package, point MDI at a folder or an inbox, or let it watch the Encompass eFolder. Every file is split into documents, each document classified, duplicates found by content, blank backs dropped, missing pages flagged. What it cannot place, a person sorts in one click.
- W-2, pay stubs, bank statements, tax returns, IDs, 1003, credit, appraisal and more
- Per-lender stacking order and eFolder names
- Batches from brokers, branches and scanners, matched to loans automatically
Every number checked
Not just extracted. Reconciled.
Every field is read with its location on the page, then checked three ways: the document's own arithmetic, the other documents in the file, and the 1003. A pay stub whose year-to-date does not add up, a bank statement with a missing page, a W-2 that disagrees with the tax return, an income figure that drifts from the application. Each finding comes with the evidence crop and a plain-language reason.
- Document math, cross-document and application checks, written as rules your team can read and edit
- Tamper signals from the PDF itself: editor fingerprints, modified-after-creation, image-only pages
- Corrections feed back into the next file
How it works
Five steps, one score.
- 1
Receive
Drop, watch folder, email-in, or the Encompass eFolder.
2Split
Packages become documents; duplicates and missing pages are flagged.
3Match
Every document lands on the right loan and the right checklist slot.
- 4
Check
Fields are extracted with evidence and reconciled across the file.
5Score
An explained score per document and per loan; only the exceptions need a person.
The score
A number you can defend in a repurchase dispute.
Every score is the sum of five parts, each with its own reasons, and every reason points at a page. Reviewers see why MDI raised a finding, which rule fired, which model read the page, and how often their team overrides that rule. Model and pipeline versions are logged on every document.
Built for the governance that Fannie Mae Lender Letter LL-2026-04 and Freddie Mac Guide section 1302.8 now expect of AI in origination: documented models, monitored accuracy, human review of what the system is not sure about.
Tamper Report
A fake can look perfect. It still has to add up.
Forensic engines inspect one page at a time, and a freshly generated pay stub gives them nothing to find. MDI checks the whole package: does the net pay land in the bank statement, does the year-to-date square with last year's W-2, does the name match across every document, was the file saved by an editor after the issuer produced it. Each document gets one of three words, Clean, Suspicious or Edited, with the evidence.
- Cross-document checks that catch what single-page forensics cannot
- Built-in PDF forensics, with Resistant AI or your own engine as a plug-in signal
- A numbered report per package, bound to the exact files by a hash, and a scorecard by submitting party
- Sold on its own to warehouse lenders, insurers and investors, by API, email-in or upload
Inside Encompass
Where the processor and the underwriter already work.
MDI connects to ICE Encompass through Developer Connect. Documents are filed to the eFolder in your stacking order, findings become conditions, and corrected figures update the application. Nothing is retyped, and the audit trail lives in both systems.
- eFolder attachments in, documents and conditions out
- Per-tenant rules, checklists and document names
- REST API and webhooks for everything else
Pricing
Priced per funded loan, not per page.
Broker
Free to start
Then $9 per loan. Drop packages, get sorted files and scores, share with your lender.
Lender
$29 per funded loan
Falling to $19 with volume. Encompass integration, custom rules and checklists, review workspace, API.
Tamper Report
Per file
For warehouse lenders and mortgage insurers. A verdict and a report for every file you fund, by API or upload. Volume pricing.
Enterprise
Annual agreement
Single-tenant deployment with your own database, bucket and encryption key, your models or ours, SOC 2 report, governance pack, uptime commitment.
Design partners wanted.
We are onboarding a small number of lenders and brokers for 2026. You get the product at cost and a say in what it checks first. We get your corrections and your opinion.
[email protected]