Adobe India Hackathon 2025 - "Connecting the Dots"

Project Overview

This repository contains solutions for the Adobe India Hackathon "Connecting the Dots" challenge, featuring intelligent PDF processing and analysis capabilities.

Challenge Structure

Challenge 1A: PDF Outline Extraction

Location: Challenge_1a/

• Extracts structured outlines (Title, H1, H2, H3 headings) from PDF documents
• Processes up to 50 pages per PDF in under 10 seconds
• Outputs hierarchical JSON format with page numbers

Challenge 1B: Persona-Driven Document Intelligence

Location: Challenge_1b/

• Analyzes collections of 3-10 related PDFs based on specific personas and job requirements
• Extracts and ranks relevant sections and subsections
• Processes document collections in under 60 seconds

Quick Start

Prerequisites

• Docker installed on your system
• Git for version control

Building and Running

Challenge 1A

# Build the Docker image
docker build --platform linux/amd64 -t challenge1a:latest Challenge_1a/

# Run the container
docker run --rm -v $(pwd)/input:/app/input -v $(pwd)/output:/app/output --network none challenge1a:latest

Challenge 1B

# Build the Docker image
docker build --platform linux/amd64 -t challenge1b:latest Challenge_1b/

# Run the container
docker run --rm -v $(pwd)/input:/app/input -v $(pwd)/output:/app/output --network none challenge1b:latest

Technical Specifications

System Requirements

• Architecture: AMD64 (x86_64)
• CPU: 8 cores, 16GB RAM
• Storage: Sufficient space for Docker images and temporary files
• Network: No internet access required during execution

Performance Constraints

• Challenge 1A: ≤10 seconds for 50-page PDFs, ≤200MB model size
• Challenge 1B: ≤60 seconds for 3-5 documents, ≤1GB model size

Project Structure

├── README.md                    # This file - Main project overview
├── Challenge_1a/               # PDF Outline Extraction
│   ├── Dockerfile              # Docker configuration
│   ├── process_pdfs.py         # Main processing script
│   ├── requirements.txt        # Python dependencies
│   ├── README.md              # Detailed documentation
│   └── sample_dataset/        # Test data and outputs
├── Challenge_1b/               # Persona-Driven Analysis
│   ├── Dockerfile              # Docker configuration
│   ├── main.py                 # Main processing script
│   ├── requirements.txt        # Python dependencies
│   ├── README.md              # Detailed documentation
│   ├── approach_explanation.md # Methodology explanation
│   └── Collection_*/          # Test collections
└── .gitignore                 # Git ignore file

Key Features

Challenge 1A Features

• Robust PDF parsing using PyMuPDF
• Intelligent heading detection with multiple strategies
• Conservative pattern matching to avoid false positives
• Hierarchical outline generation with page numbers
• Fully offline processing

Challenge 1B Features

• Semantic similarity analysis using sentence transformers
• Persona-driven content relevance scoring
• Multi-document collection processing
• Intelligent section ranking and extraction
• Comprehensive metadata tracking

Compliance

• ✅ AMD64 Docker compatibility
• ✅ No internet access required
• ✅ CPU-only processing (no GPU dependencies)
• ✅ Model size constraints met
• ✅ Runtime performance requirements satisfied
• ✅ Modular, reusable code architecture

Documentation

Each challenge directory contains detailed documentation:

• Challenge_1a/README.md: Complete implementation guide
• Challenge_1b/README.md: Comprehensive usage instructions
• Challenge_1b/approach_explanation.md: Methodology explanation

Testing

Both challenges include comprehensive test datasets:

• Challenge_1a: 6 sample PDFs with expected outputs
• Challenge_1b: 3 diverse collections (Travel, Adobe Learning, Recipes)

Contact

For questions or issues, please refer to the individual challenge documentation or contact the project maintainer.

---

Note: This repository should remain private until the competition deadline as per hackathon guidelines.