Introduction to OpenDataLoader

OpenDataLoader PDF is an open-source PDF parser that ranks #1 in extraction accuracy across independent benchmarks. It extracts structured data from PDFs with bounding boxes, making it perfect for RAG pipelines, document analysis, and AI applications.

Why OpenDataLoader?

Benchmark Results

| Parser                  | Overall | Tables | Speed    |
|------------------------|---------|--------|----------|
| OpenDataLoader [hybrid] | 0.907   | 0.928  | 0.46s/pg |
| docling                 | 0.882   | 0.887  | 0.76s/pg |
| marker                  | 0.861   | 0.808  | 53.9s/pg |
| pymupdf4llm            | 0.732   | 0.401  | 0.09s/pg |

Installation

Step 1: Verify Java

OpenDataLoader requires Java 11 or higher:

java -version

If not installed, download from [Adoptium](https://adoptium.net/).

Step 2: Install the Python SDK

pip install -U opendataloader-pdf

Your First Conversion

Create a simple Python script:

import opendataloader_pdf

# Convert a PDF to Markdown and JSON
opendataloader_pdf.convert(
    input_path="document.pdf",
    output_dir="output/",
    format="markdown,json"
)

This creates two files in output/:

• document.md - Clean Markdown text
• document.json - Structured data with coordinates

Output Formats

OpenDataLoader supports multiple output formats:

Combining Formats

opendataloader_pdf.convert(
    input_path="document.pdf",
    output_dir="output/",
    format="markdown,json,html"  # Multiple formats
)

Understanding JSON Output

The JSON output contains rich metadata:

{
  "file name": "document.pdf",
  "number of pages": 5,
  "kids": [
    {
      "type": "heading",
      "level": 1,
      "page number": 1,
      "bounding box": [72.0, 700.0, 540.0, 730.0],
      "content": "Introduction"
    },
    {
      "type": "paragraph",
      "page number": 1,
      "bounding box": [72.0, 650.0, 540.0, 690.0],
      "content": "This is the first paragraph..."
    }
  ]
}

Key Fields

• type: Element type (heading, paragraph, table, list, image)
• bounding box: [left, bottom, right, top] in PDF points
• page number: 1-indexed page reference
• content: Extracted text

Challenge: Element Counter

Write a script that:

• Converts a PDF to JSON
• Counts elements by type (headings, paragraphs, tables, etc.)
• Prints a summary report

import opendataloader_pdf
import json

# Your code here:
# 1. Convert a sample PDF
# 2. Load the JSON output
# 3. Count elements by type
# 4. Print results

Expected output:

Element Summary:
- headings: 12
- paragraphs: 45
- tables: 3
- images: 2
Total elements: 62