PDF Analysis

Overview

Turns PDFs into answers: summaries with page citations, structured field extraction, and clean tables — while being explicit about pages that couldn't be read (scans, images, broken encoding).

When to Use

• "Summarise this PDF", "what does page 12 say about…", "pull the totals from these invoices"
• Extracting tables to CSV, comparing two versions of a contract or report.

Usage

1. Inspect first: page count, whether text is extractable, and which pages are scanned images. With Python:

   import pdfplumber
   with pdfplumber.open(path) as pdf:
       for i, page in enumerate(pdf.pages, 1):
           text = page.extract_text() or ""
           print(i, len(text))  # ~0 chars → likely scanned; needs OCR

2. Extract text per page and keep page numbers attached to every chunk.
3. Tables: use page.extract_tables(); check headers and merged cells by eye against the page; export to CSV if asked.
4. Field extraction (invoices, forms, statements): define the fields up front, extract each with its page number, and return JSON. Leave a field null rather than guessing.
5. Answer or summarise with citations like (p. 7). For long documents summarise per section first, then overall.
6. Report gaps: list pages that were scanned/unreadable and whether OCR was used.

Examples

Prompt: "Get invoice number, date, vendor and total from these 20 PDFs into a CSV."

See examples/invoice-fields.md for the field spec and output format.

Requirements

• File access to the PDF. Python with pdfplumber (or pypdf) improves table extraction; OCR needs tesseract.

Limitations

• Scanned PDFs need OCR, which introduces errors — numbers from OCR should be double-checked.
• Complex layouts (multi-column, rotated tables) may extract out of order.
• Password-protected PDFs require the user to unlock them first.

Changelog

• 1.0.0 — Initial release.