# Document Intelligence

## Everydocument.One engine.

Curvestone extracts structured data from every document type in a mortgage case file — payslips, bank statements, tax returns, and open banking feeds — through one unified pipeline.

### Curvestone AI

## Steps

1. Ingest document  
2. Classify  
3. Extract fields  
4. Normalise

### Input: payslip_nov.jpg  
### Extract fields

**User Input · List items**  
1. Gross pay · YTD total  
2. Employer name · tax code  
3. NI number · pay period

## Capabilities

### Messy inputs. Clean, structured output.

### Multi-modal extraction

PDFs, scanned images, and native digital documents — all handled by the same extraction pipeline. No parallel systems.

### Payslips & employment

Gross income, net pay, employer name, deductions, frequency — extracted and normalised automatically.

### Bank statements

Identifies salary credits, recurring commitments, and unusual patterns across 3–12 months of statements.

### Tax returns & accounts

SA302s, tax year overviews, company accounts — structured into comparable data points for affordability.

### Open banking feeds

Structured open banking data consumed through the same engine as document inputs — no parallel pipelines.

### Credit bureau data

Ingests Experian, Equifax, and TransUnion data for a unified affordability picture alongside documents.

## Coverage

> “Payslip photographs that would defeat any generic OCR. Extracted, normalised, and verified — in under three seconds.”

200+ Document types supported natively  
<3s Per document extraction  
99% Field-level accuracy

## How it works

### From scanned chaos to a single structured schema.

01

### Any format, any source

Photographed payslips. Password-protected PDFs. Handwritten forms. Open banking JSON. The engine handles all of it through a single upload.

02

### Extraction & normalisation

Twelve algorithms working in concert — LLMs, OCR, vision models — extract fields and normalise them into a consistent structured schema.

03

### Validated output

Extracted data is cross-referenced against other documents in the case file. Inconsistencies flagged. Clean data delivered into your workflow.

## 200+ document types.One extraction engine.

See Document Intelligence in action on your case file formats.