Gemini 3.5 Flash · full-document mode

pdf-inspector

Decompile entire PDFs into real HTML — every page rendered, stitched into one continuous-flow document. Every img, table, p is a real DOM node. $$('h1') returns headings across all pages.

Powered by Gemini 3.5 Flash one-shot rebuild + lossless extract_image(xref) via PyMuPDF + 600 DPI vector fallback. ~20-50 seconds per page, 3 pages in parallel. ~$0.005-$0.01 per page.

4 full PDFs · click a card to open the inspector

3M 2025 Global Impact Report

3M 2025 Global Impact Report

Corporate sustainability report · 56 pp · figures, charts, scope icons

56 pages · continuous scroll
CHAI 2025 Roadmap

CHAI 2025 Roadmap

Startup deck · 8 pp · charts + GPU board photos

8 pages · continuous scroll
Attention Is All You Need

Attention Is All You Need

Academic paper · 15 pp · authors, abstract, model architecture

15 pages · continuous scroll
Synthetic invoice

Invoice (synthetic)

Line-item table · 1 pp · structured form

1 page · text-only

Try a recipe · transform the demo page via ?init=

Whole-doc stats (3M, 56 pp)

$$('section.pdf-page').length + ' pages, ' + $$('h1,h2,h3').length + ' headings'

Pull author emails (Attention)

$$('.author').map(a => ({name, email}))

Extract invoice line items

$$('tbody tr').map(r => ({SKU, Description, Qty, Unit, Amount}))

Image alt-text catalog (3M, all pages)

$$('img').map(i => i.alt) → every figure across the report

Build a table of contents (CHAI)

$$('h1,h2,h3').map(h => h.tagName + ' ' + h.textContent.trim())

Count paragraphs (Attention, 15 pp)

$$('p').length + ' paragraphs across all pages'