Adobe-Hackathon-2025-HackerFem

Round 1A - Document Outline Extractor

This project is a solution for Round 1A of the Adobe India Hackathon, "Connecting the Dots." It's designed to parse a PDF file and automatically extract a structured outline, including the document's title and hierarchical headings (H1, H2, H3).

---

Our Approach

The solution processes PDFs by analyzing their structural properties, primarily focusing on font characteristics.

1. Text Extraction: The PyMuPDF library is used to extract all text blocks from each page while preserving crucial metadata like font size.
2. Noise Reduction: A series of cleaning steps are applied to remove irrelevant text fragments and whitespace, ensuring that only meaningful content is analyzed.
3. Heading Identification: The core of the logic assumes that headings use larger font sizes than body text. The program identifies all unique font sizes in the document and maps the three largest sizes to H1, H2, and H3 respectively.
4. Bonus: Language Detection: To handle the multilingual requirement, the langdetect library is used to determine the primary language of the document, which is included in the final output.

---

Libraries Used

This solution is lightweight and relies on two key Python libraries:

• PyMuPDF: Used for robust PDF parsing and text extraction.
• langdetect: Used for the multilingual heading detection bonus criterion.

---

How to Build and Run

The solution is containerized using Docker and is designed to run offline on a CPU.

1. Build the Docker Image • From the root directory of the project, run the following command to build the image:

docker build --platform linux/amd64 -t mysolutionname:somerandomidentifier .

2. Prepare Input • Place your PDFs inside a folder named 'input' in the root directory.

3. Run the Container • The container will process all .pdf files in the input folder and produce corresponding .json files in the output folder:

docker run --rm -v $(pwd)/input:/app/input -v $(pwd)/output:/app/output --network none mysolutionname:somerandomidentifier

---

Example Output • For a sample PDF sample.pdf, the output JSON in output/sample.json looks like:

{ "title": "Sample Document", "language": "en", "outline": [ { "level": "H1", "text": "Introduction", "page": 1, "language": "en" }, { "level": "H2", "text": "Overview", "page": 2, "language": "en" } ] }

Notes • Runs in <10 seconds for a 50-page PDF. • No internet access required.