99% accuracy document extraction with human review.
- Claude AI Vision — primary extraction engine, reads vector PDFs, handwriting, stamps
- Structured text output — headings, label:value pairs, paragraphs all classified
- Fully editable results — click any cell, field, or text to correct it
- Light theme — clean professional UI
- Human review mode — toggle edit mode, fix any error, save changes
- Python 3.10 or 3.11 (NOT 3.14 — use 3.11 for best compatibility)
- Node.js 18+
- Tesseract: https://github.com/UB-Mannheim/tesseract/wiki → install to
C:\Program Files\Tesseract-OCR\
The main.py includes a compatibility patch for the pkgutil.find_loader removal.
cd docparser_v3\backend
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
python -m uvicorn main:app --reload --port 8000cd docparser_v3\frontend
npm install
npm start- Claude Vision API (primary) — analyzes the rendered page image, understands layout, tables, handwriting. Returns structured JSON.
- PyMuPDF native (for real-text PDFs) — perfect extraction when text is embedded.
- Tesseract OCR (fallback) — used when Claude API is unavailable.
- Human Review — edit mode lets you fix any remaining errors inline.
| Error | Fix |
|---|---|
pkgutil.find_loader ImportError |
Already patched in main.py |
pymupdf build fails |
pip install pymupdf==1.27.2.3 (pre-built wheel) |
| Tables empty | Install Tesseract + restart backend |
| Claude API 401 | API key is injected automatically in the artifact context |