# The whole loan package: index, extract, clear and QC.

*Document Intelligence*

> Sei reads every document in the package, calculates income with rep-and-warrant relief, clears conditions against agency guidelines plus your overlays, automates the Closing Disclosure and QCs every loan — citing the source page on every finding. Your underwriters see only the exceptions.

## The desks Document Intelligence powers

- **[POS](/for-lenders/pos)** — The borrower’s front door, inside the system
- **[Indexing Agent](/for-lenders/indexing-agent)** — Every page split, named and filed the moment it lands
- **[Document Concierge](/for-lenders/document-concierge)** — Gets the full document set before underwriting asks
- **[Income Desk](/for-lenders/income-desk)** — Rep-and-warrant-eligible income, cited to the page
- **[Conditions Manager](/for-lenders/conditions-manager)** — Creates conditions against the guide, then clears them
- **[TRID & Cure Management](/for-lenders/trid-and-cure-management)** — Watches every clock, catches every cure
- **[Pre-Close QC Agent](/for-lenders/pre-close-qc-agent)** — Every file before it funds, not a ten percent sample
- **[Loan Boarding Agent](/for-servicers/loan-boarding-agent)** — Catches boarding defects before the first statement
- **[Indexing Manager](/for-servicers/indexing-manager)** — Every letter, fax and upload filed and routed the day it lands
- **[Escrow Manager](/for-servicers/escrow-manager)** — Analysis, shortages, and the explanation that is borrower-friendly
- **[Default Management Desk](/for-servicers/default-management-desk)** — Sixty days and beyond, with the timelines held
- **[Loss Mitigation Agent](/for-servicers/loss-mitigation-agent)** — Complete packages, evaluated inside the Reg X clock
- **[Claims Analyst](/for-servicers/claims-analyst)** — The claim package, read and reconciled for review
- **[Payoffs Manager](/for-servicers/payoffs-manager)** — Quotes, good-through dates and the release clock

## A package in, a cleared file out

1. **Ingest** — The whole package is classified and assembled by agentic OCR, including the scanned and non-standard layouts that defeat template extraction.
2. **Calculate and validate** — Income is calculated across W-2, self-employed, rental and retirement sources. Assets are verified and discrepancies surfaced against the agency handbooks and your overlays.
3. **Clear, close and QC** — Conditions clear with the rule and the source page cited, written back into your LOS. The Closing Disclosure is checked for TRID timing and fee tolerances, and every loan is QC’d before funding and after close.

## Proof

Better Mortgage, the NASDAQ-listed digital lender, moved QC onto Sei.

- **4h 50m → 47m** — loan review time
- **92%** — of conditions checks automated
- **78%** — items auto-cleared

Loan review fell from **4 hours 50 minutes to 47 minutes** per file, and 78% of items now clear at high confidence — so QC touches only the exceptions.

[Read the case study](/case-study/sei-x-nasdaq-lender)

## Demo

[Document Intelligence on a loan file](https://www.youtube.com/watch?v=1nGSGhXZZVU) — Intake through condition clearing on a real package, with the citation behind each finding.

## The documents, the loan types and the guidelines

### Documents it reads

- Pay stubs, W-2s and 1099s
- Federal returns, Schedule C, K-1 and S-corp filings
- Bank and asset statements
- Verifications of employment and deposit
- Appraisals and the Submission Summary Report
- Title commitments and insurance binders
- Scanned and non-standard layouts that break template extraction

### Loan types

- Conventional
- FHA
- VA
- USDA
- Jumbo
- HELOC
- Non-QM and other non-agency

### Rule sets it clears against

- Fannie Mae Selling Guide
- Freddie Mac Seller/Servicer Guide
- FHA Handbook 4000.1
- Your investor overlays and credit policy
- TRID timing and the fee-tolerance buckets

## What sets it apart

- Income calculated with representation-and-warranty relief
- Conditions cleared against the agency handbooks and your own overlays, each item confidence-scored
- Every finding cites the document and the page it came from, so a reviewer checks the citation rather than the answer
- Closing Disclosure automation and post-close QC on the same platform that underwrote the loan
- Results written back into your LOS

**47 min** — loan QC per file — down from 4h 50m

## When it is unsure

Any finding below your confidence threshold routes to your own underwriter as a named exception, with the rule, the reasoning and the source page attached.

## FAQ

**Does Sei replace our LOS?**

No. Sei sits above the loan origination system you already run — ICE Encompass, Calyx and MeridianLink among them — and writes its results back into it. There is no rip-and-replace, and your system of record stays your system of record.

**What does rep-and-warrant relief actually depend on?**

Income calculated through Fannie Mae’s Income Calculator. Sei runs the calculation there across W-2, self-employed, rental and retirement income, so eligible calculations earn representation-and-warranty relief on the income component and lower your repurchase risk.

**How do we know why a condition cleared?**

Every decision is logged with the rule it was made against, the reasoning that produced it, a confidence score and a citation to the source document and page. Months later, the question “why did it clear this?” still has an answer you can hand to an auditor.

**Can it work against our overlays, not just the agency guides?**

That is the point of it. Your investor overlays, underwriting guidelines, credit policy and QA rubrics are encoded alongside the Fannie, Freddie and FHA handbooks. The agent reasons against the combined set, rather than against a general model’s idea of mortgage lending.

## How we think about AI agents in underwriting

- [Calculating Self-Employed Income With an AI Agent: The 1084 Cash-Flow Analysis, the 4506-C Transcript, and the Reasonableness Call the Underwriter Owns](/blog/ai-income-calculation-self-employed-1084-4506c-reasonableness) — Self-employed income is where mortgage underwriting is slowest, most inconsistent, and most exposed to fair-lending risk, because two underwriters can read the same tax returns and reach different qualifying income. An AI agent can run the Form 1084 cash-flow analysis the same way every time and document every add-back to its line on the return. What it cannot do is make the reasonableness determination Fannie assigns to the underwriter, and building the agent so it stops at that line is the whole design problem.
- [Conditions Clearing Is the Slowest Part of the Loan, and the Best Place to Put an AI Agent](/blog/ai-loan-processing-conditions-clearing-los-reg-b-incompleteness) — The gap between conditional approval and clear-to-close is where cycle time goes to die. How an AI processing agent clears conditions inside the LOS without breaking the Reg B incompleteness clock or the TRID redisclosure rules.
- [The Three Tolerance Buckets: Balancing the Closing Disclosure Against the Loan Estimate With an AI Agent](/blog/trid-fee-tolerances-closing-disclosure-cures-ai-mortgage) — Between the Loan Estimate and the Closing Disclosure sits the fee comparison that decides whether a lender owes the borrower a refund. Zero tolerance, ten percent aggregate, and no tolerance are three different rules on three different sets of fees, and a changed-circumstance re-disclosure can move a fee from one bucket to another. Where an AI agent tracks every fee from LE to CD, catches the tolerance breach before consummation, and computes the cure the rule requires.

---

_Source: [https://seiright.com/platform/document-intelligence](https://seiright.com/platform/document-intelligence) · Sei AI_