Skip to content

README.md

Latest commit

 

History

History
139 lines (104 loc) · 3.43 KB

README.md

File metadata and controls

139 lines (104 loc) · 3.43 KB

📄 PDF Processing Solution — Adobe India Hackathon 2025

🧠 Overview

This solution processes PDF files and extracts a structured outline based on headings. It outputs the result as a JSON file per PDF and adheres strictly to the challenge constraints:

  • No machine learning
  • Fully offline
  • Sub-10s processing for 50-page PDFs
  • ≤ 16GB RAM
  • Works on Linux (AMD64)

🧰 Approach

This is a logic-based, non-ML solution that:

  1. Extracts raw text + font/position data from PDFs using pdf.js-extract

  2. Converts content into markdown-style blocks, assigning heading levels (H1–H3) based on:

    • Font size
    • Font weight (bold/italic)
    • Vertical position
    • Regex-based heuristics
    • Text block analysis
    • Visual/positional logic

Template heading filtering

  1. Identifies headings from markdown and tags them with page numbers

  2. Applies heuristic filtering to remove:

    • Repetitive headers (e.g. "Page 1", "Confidential")
    • Common footer patterns

  3. Outputs a JSON structure matching the provided schema:

    {
  "title": "Document Title",
  "outline": [
    {
      "level": "H1",
      "text": "Heading Text",
      "page": 1
    },
     {
      "level": "H2",
      "text": "Heading Text",
      "page": 2
    },
     {
      "level": "H3",
      "text": "Heading Text",
      "page": 3
    }
  ]
}

No external APIs or internet access are used at any point.

📦 Libraries & Tools Used

All tools are open-source and installed inside the Docker container:

Library Purpose
pdf.js-extract Extracts structured data from PDFs
marked Markdown parser for heading extraction
Node.js (v18) Main runtime for logic

📁 Project Structure

Adobe_Challange_1_a/
├── input/                      # 📥 Place your input PDFs here
├── output/                     # 📤 JSON outputs appear here
├── sample_dataset/             # 🧪 Sample input/output/testing
│   ├── pdfs/
│   ├── outputs/
│   └── schema/output_schema.json
├── Dockerfile                  # 🐳 Docker config (in root)
├── process_pdfs.js             # 🚀 Main logic
├── package.json                # 📦 Node.js dependencies
└── README.md                   # 📚 You're reading it

🛠️ How to Build & Run

🔒 Your solution must work in an isolated, offline, CPU-only Docker container.

▶️ 1. Clone & Build Docker Image

git clone https://github.com/harshcodesdev/Adobe_Challange_1_a.git
cd Adobe_Challange_1_a

docker build --platform linux/amd64 -t pdf-processor .

🧪 2. Run with Sample Data (Test Run)

docker run --rm \
  -v $(pwd)/sample_dataset/pdfs:/app/input:ro \
  -v $(pwd)/sample_dataset/outputs:/app/output \
  --network none \
  pdf-processor

➡ Output will appear in: sample_dataset/outputs/

📁 3. Run with Your Own Data

  1. Place your PDFs in the input/ folder
  2. Make sure output/ folder exists and is writable
  3. Then run:
docker run --rm \
  -v $(pwd)/input:/app/input:ro \
  -v $(pwd)/output:/app/output \
  --network none \
  pdf-processor

➡ For every filename.pdf, you’ll get filename.json in output/