🧠 Adobe Hackathon 2025 – Round 1B

Persona-Based Intelligent Section Extractor for PDFs

📌 Problem Statement

Develop a document intelligence system that:

• Accepts user-defined folders (collections) of PDFs.
• Reads a persona and job-to-be-done from a JSON file.
• Extracts and ranks the most relevant sections based on the persona/job.
• Runs fully offline and is Dockerized.

---

✅ Solution Overview

This system:

• Iterates over every CollectionX folder present (e.g., Collection1, Collection2, ...).
• Parses all PDFs inside the PDFs/ subfolder.
• Extracts titles/headings/subsections that are contextually relevant.
• Ranks sections and outputs results to challenge1b_output.json.

---

📁 Folder Structure

project_root/
├── main.py                         # Entry point — runs processing for all collections
├── src/
│   └── processor.py                # Persona-based extraction and ranking logic
├── Collection1/
│   ├── PDFs/                       # Folder containing input PDFs
│   ├── challenge1b_input.json      # Persona & job definition
│   └── challenge1b_output.json     # Output written here after processing
├── Collection2/
│   ├── PDFs/
│   ├── challenge1b_input.json
│   └── challenge1b_output.json
├── Dockerfile                      # Offline Docker image
├── requirements.txt                # Python dependencies
├── README.md                       # This file

✅ You can create as many collections as needed — the script auto-detects and processes each one.

---

🧠 Input JSON Format — challenge1b_input.json

{
  "persona": "Investment Analyst",
  "job": "Analyze revenue trends, R&D investments, and market positioning strategies"
}

---

📤 Output JSON Format — challenge1b_output.json

{
  "metadata": {
    "input_documents": ["report1.pdf", "report2.pdf"],
    "persona": "Investment Analyst",
    "job_to_be_done": "Analyze revenue trends, R&D investments, and market positioning strategies",
    "processing_timestamp": "2025-07-15T12:00:00Z"
  },
  "extracted_sections": [
    {
      "document": "report1.pdf",
      "page_number": 3,
      "section_title": "Revenue Growth Breakdown",
      "importance_rank": 1
    }
  ],
  "subsection_analysis": [
    {
      "document": "report1.pdf",
      "page_number": 3,
      "refined_text": "Revenue increased due to product diversification..."
    }
  ]
}

---

🐳 Running the Project with Docker (Offline)

Step 1: Build the Docker Image

docker build -t persona-section-extractor .

Step 2: Run the Container

✅ On Linux/macOS (bash, zsh):

docker run --rm -v $(pwd):/app --network none persona-section-extractor

✅ On Windows (PowerShell):

docker run --rm -v "${PWD}:/app" --network none persona-section-extractor

✅ This command:

• Mounts the current project directory into the container.
• Disables all internet access (--network none).
• Automatically processes all CollectionX/ folders (e.g., Collection1, Collection2, ...).
• Writes output JSONs back into each respective collection directory.

---

⚙️ Dependencies

Everything is already included in the Docker image.

If you want to install dependencies locally:

pip install -r requirements.txt

---

🧪 Testing Your Setup

1. Add your own CollectionX/ folder (e.g., Collection4/).
2. Add a few PDFs inside the PDFs/ subfolder.
3. Add a challenge1b_input.json defining the persona/job.
4. Run the Docker container again.
5. Check challenge1b_output.json for results.

---

🧾 Example Collections

Collection1/      # Travel Planning
├── PDFs/
├── challenge1b_input.json
├── challenge1b_output.json

Collection2/      # Adobe Acrobat Learning
├── PDFs/
├── challenge1b_input.json
├── challenge1b_output.json

Collection3/      # Recipes
├── PDFs/
├── challenge1b_input.json
├── challenge1b_output.json

---

👩‍💻 Author

Salusha — Participant, Adobe Hackathon 2025
GitHub: https://github.com/salusha

Snehal Taori — Participant, Adobe Hackathon 2025
GitHub: https://github.com/snehaltaori

Deepanshi Verma — Participant, Adobe Hackathon 2025
GitHub: https://github.com/DeepanshiiVerma

---