AI PDF Reader Demo

📄

Drop a PDF here or click to browse
Text is extracted locally — only your question goes to OpenAI

Extracted Text —

Extracted text will appear here after loading a PDF…

AI Chat Strategy 1 — PDF.js + GPT-4o-mini

Load a PDF above then ask questions about it.

📄

Drop a PDF to extract structured fields
Invoice · Resume / CV · Contract

Schema Selection

Document type

Fields for selected schema

Extracted JSON

// Upload a PDF and click Extract // Returns a typed JavaScript object // No JSON.parse needed — guaranteed schema

Which strategy should I use?
Check if you can select text in the PDF → Strategy 1.
Scanned or image PDF → Strategy 2.
Need typed fields (invoice, resume, contract) → Strategy 3.

PDF.js Text Extraction + Chat ~$0.0002/page

Extract text with PDF.js → clean it → send to gpt-4o-mini as context → chat Q&A.

✅ Digital PDFs with selectable text ✅ Reports, articles, e-books — up to ~40 pages ✅ Cheapest and fastest option ❌ Fails on scanned / image-only PDFs

Token efficiency after cleaning

GPT-4o Vision (Page as Image) ~$0.01–0.02/page

Render PDF pages to canvas → export as base64 PNG → send to GPT-4o Vision API.

✅ Scanned PDFs and image-only documents ✅ Complex layouts, charts, handwritten notes ✅ Works on any PDF type regardless of text encoding ❌ ~50× more expensive than Strategy 1 ❌ Slow — must render and upload each page

Cost per page relative to Strategy 1

Structured JSON Extraction ~$0.0003/page

Extract text → define a JSON schema → AI fills in the fields → typed object returned. Uses response_format: json_object.

✅ Invoices, receipts, bills ✅ Resumes and CVs ✅ Contracts and agreements ✅ Any document with predictable fields ❌ Not for open-ended Q&A

Unique to this tutorial — not covered by competitors

Token waste warning: Raw PDF extraction wastes 40–60% of tokens on layout artifacts (page numbers, headers, footers, separator lines). Always run cleanPdfText() before sending to the AI — it typically cuts token count nearly in half with zero quality loss.