Extracting data from loan applications
Nearly every US residential mortgage starts with a Uniform Residential Loan Application, Fannie Mae Form 1003 or Freddie Mac Form 65. The lender's loan origination software fills it in. Processors, underwriters and loan buyers review the signed copy for every loan, as a PDF in the loan file.
Underwriting needs the loan amount and purpose, the property value and occupancy, income, assets and liabilities. Those feed automated underwriting, which checks them against the borrower's documents and returns the findings that set the loan's conditions.
What makes loan applications hard
Selections carry the meaning
Loan purpose, occupancy, citizenship and every declaration are radio buttons or checkboxes, so the answer is a mark beside a printed option.
Arithmetic across pages
Liabilities listed on one page feed the funds-to-close calculation on another, so related figures sit several pages apart.
System values over a form face
A loan origination system prints its values in its own typeface over the form's lines, often a little off the line they belong to.
Parse, extract, verify
Parse
Parse turns the document into reading-order Markdown and a hierarchy of blocks (text, tables, figures), and grounds each block to its exact location on the page.
Extract
You define the fields in a JSON schema, and Extract returns their values in a consistent, structured format, each linked back to where it appears in the Markdown.
Verify
Each value links back to where it appears, down to the character range in the Markdown, and the page and bounding box in the source document, so every result can be verified.
One uniform residential loan application, end to end
About this document: A completed, e-signed Form 1003 for a single-borrower cash-out refinance, generated by a lender's document system. Eight portrait pages: the borrower sections on pages 1 to 6, then the lender's loan information on pages 7 and 8. Page 3 is featured. Every personal value and signature was replaced with stand-ins before parsing.
Processing objective: The objective is to read the loan being applied for: how much, for what purpose, against what property value, and how the borrower will occupy the home. Purpose and occupancy are ticked boxes, and each value grounds to its own line of the form.
Demonstration fields: Loan amount, Loan purpose, Property value, Occupancy, Number of units
The document
4 of 8 pages
Page 1
Page 2
Page 3shown below
Page 7
Where each value came from
loan_and_property.loan_amount
{
"value": 404500,
"page": 3,
"range": {
"end": 10049,
"start": 10023
},
"box": {
"xmin": 0.02554,
"ymin": 0.38216,
"xmax": 0.97421,
"ymax": 0.39853
}
}loan_and_property.loan_purpose
{
"value": "Refinance",
"page": 3,
"range": {
"end": 10095,
"start": 10080
},
"box": {
"xmin": 0.02554,
"ymin": 0.38216,
"xmax": 0.97421,
"ymax": 0.39853
}
}loan_and_property.property_value
{
"value": 608200,
"page": 3,
"range": {
"end": 10267,
"start": 10238
},
"box": {
"xmin": 0.15202,
"ymin": 0.4327,
"xmax": 0.62661,
"ymax": 0.44852
}
}loan_and_property.occupancy
{
"value": "Primary Residence",
"page": 3,
"range": {
"end": 10302,
"start": 10279
},
"box": {
"xmin": 0.02509,
"ymin": 0.44952,
"xmax": 0.91703,
"ymax": 0.4659
}
}loan_and_property.number_of_units
{
"value": 1,
"page": 3,
"range": {
"end": 10237,
"start": 10219
},
"box": {
"xmin": 0.15202,
"ymin": 0.4327,
"xmax": 0.62661,
"ymax": 0.44852
}
}How to try it yourself
Everything on this page is reproducible. The sample, its schema, and the Parse and Extract API outputs are published. You can repeat the same two calls from whichever surface you already build on.
- Claude Skilldescribe the job; the agent writes it
- CLIno code at all
- MCPexpose ADE docs as an agent tool
- REST APIany language
- Pythonlandingai-ade
- TypeScriptlandingai-ade
Documents into data
Accurate
The best accuracy on the difficult documents other tools get wrong, such as messy scans, complicated tables, checkboxes and handwriting.
Auditable
Every answer cites its sources. Values come back with an audit trail (page, bounding box, character range) so a human can review them.
Agentic
Applies multi-step reasoning, and plans before it acts. Handles documents it has never seen before and adapts to variations.
API-first
Composable APIs that drop into any tech stack and can be called from anywhere. Easy to build on and embed in your pipelines and products.
Affordable
Cost-competitive rates with complete price transparency. Spend controls before you run, and every job reports its exact cost.
Enterprise-grade
Built for production and scale from day one. Fully observable and monitored, running securely in our cloud or yours.
Built for high-volume, high-throughput production
- Zero data retention
- Priority rate limits
- Enhanced support
- VPC and on-prem deployment
Available on Team and Enterprise plans.
