pdfduck

bulk pdfs to structured csv

---

what it does

drop PDFs (shipping bills, invoices, BoL, any tabular documents) and get clean, structured CSV output.

backend extracts tables using pdfplumber. frontend displays results and downloads CSV. zero AI API costs. no gemini, no openai — pure python pdf parsing.

---

architecture

Frontend (Cloudflare Pages)
  → uploads PDF via POST
  → displays preview table
  → downloads CSV

Backend (Railway - Python FastAPI)
  → receives PDF
  → extracts tables with pdfplumber
  → returns structured JSON
  → falls back to text pattern matching if no tables

---

how it works

extraction logic:

1. pdfplumber opens the PDF and scans for tables
2. first row of each table = column headers
3. remaining rows = data rows
4. all rows from all tables are merged into a single JSON array
5. if no tables found, it falls back to regex pattern matching for common fields:
   • invoice number, date, amounts, consignee, shipper, email, phone, etc.
6. if still minimal data, it extracts line-by-line (useful for resumes, general docs)

what works best:

• PDFs with clear tabular structure (bordered or grid-based tables)
• invoices, shipping bills, BoL with line-item tables
• multi-page documents with consistent table formats

what doesn't work:

• heavily scanned/skewed PDFs (poor OCR quality)
• PDFs with complex multi-column layouts that aren't tables
• pure image PDFs without embedded text

---

project structure

.
├── backend/
│   ├── main.py              
│   ├── requirements.txt     
│   ├── Procfile (optional)
│   ├── railway.json         
│   └── README.md            
│
├── index.html               
├── favicon.ico              
├── build.sh (injects env vars)
└── README.md

---

features

✓ bulk upload (multiple PDFs at once)
✓ per-file status tracking (pending → reading → done/error)
✓ live progress bar
✓ preview table before download
✓ CSV generation (client-side, zero server storage)
✓ copy to clipboard
✓ mobile responsive
✓ no AI API costs — pure PDF parsing
✓ secure backend URL via environment variables

---

api reference

POST /extract

upload a PDF, get structured JSON back.

request:

curl -X POST http://your-backend/extract \
  -F "file=@document.pdf"

response (success - tables found):

{
  "success": true,
  "method": "table_extraction",
  "rows": 15,
  "data": [
    {
      "item": "Widget A",
      "quantity": "100",
      "price": "50.00",
      "_source_page": 1,
      "_source_table": 1,
      "_source_row": 1
    }
  ]
}

response (success - text extraction):

{
  "success": true,
  "method": "text_extraction",
  "rows": 1,
  "data": [
    {
      "name": "John Doe",
      "email": "john@example.com",
      "phone": "+1-555-0100",
      "line_1": "Software Engineer",
      "_full_text": "...",
      "_text_length": 1523
    }
  ]
}

---

contributing

open source. do what you want. PRs welcome.