PDF batch extractor

Extract structured data from multiple PDFs into Excel.

When to use it

Documents with a fairly fixed layout: invoices, customs forms, lab reports, statements. Free-form documents need layout analysis first, which costs far more.

Flow

Steps

1. Scan — walk the directory, filtering by extension
2. Split — text-layer PDFs are extracted directly; scans go through OCR
3. Locate fields — by template: a keyword anchor plus a relative position
4. Validate — check amount, date and tax-id formats; normalise units
5. Write out — one row per document, invalid rows flagged separately

Options

Option	Description	Default
input_dir	Input directory	required
template	Field template name	required
ocr	OCR when no text layer exists	true
output	Output Excel path	./out.xlsx
on_invalid	On validation failure: flag | skip	flag

Field template

fields:
  - name: invoice_no
    anchor: Invoice No.
    offset: right
    pattern: '^\d{8,20}$'
  - name: amount
    anchor: Total
    offset: right
    type: currency

The limits of OCR accuracy

On scans, digit errors cluster on 0/O, 1/l and 5/S. Always enable format validation for amount fields, and keep a link to the source image on flagged rows — silently recording a wrong number is far more dangerous than failing loudly.