PDF Document Text Extraction using Amazon Bedrock

This project demonstrates how to extract and compare text from PDF documents using two powerful image-capable Large Language Models (LLMs) available through Amazon Bedrock:

• Claude 3.7 Sonnet by Anthropic
• Amazon Nova Pro by Amazon

Each page of a PDF is converted into an image and sent to both models in parallel to extract textual information. The results are saved as .txt files per page for further analysis or use.

📁 Project Structure

├── documents.ipynb   # Original notebook
├── documents/        # Folder where input PDF files are stored
├── images/           # Folder where images of PDF are stored
├── texts/            # Folder where extracted text results are saved
└── README.md         # This file

🔧 How It Works

Step 1: Convert PDF Pages to Images

Using the fitz library, the notebook converts each page of a PDF into high-quality PNG images.

Step 2: Define LLM Extractor Functions

Two asynchronous Python functions are defined to:

• Encode the image in base64
• Send it to each model via boto3 and the Amazon Bedrock API
• Parse and return the extracted text response

Step 3: Parallel Processing

Using asyncio.gather, the notebook sends the same image to both models concurrently for faster processing.

Step 4: Save Results

The extracted text is saved as .txt files per page with side-by-side comparisons of both models.

📊 LLM Metrics Tracked

For each model, the following metrics are logged and printed:

Claude 3.7 Sonnet:

• Input Tokens: 1666
• Output Tokens: 1036
• Start Time: 1746124263.978323
• End Time: 1746124292.999416

Amazon Nova Pro:

• Input Tokens: 2223
• Output Tokens: 971
• Start Time: 1746124263.98382
• End Time: 1746124279.478841

📦 Requirements

• Python 3.11+
• AWS credentials with access to Amazon Bedrock
• Dependencies:

  pip install boto3 pymupdf

🚀 How to Run

1. Upload a PDF to the documents/ folder.
2. Open and run the documents.ipynb notebook.
3. Extracted text files will be saved in the texts/ folder.

🤝 Contributing

Feel free to fork this repo and improve model selection, add post-processing like IdP comparison, or integrate evaluation metrics like BLEU or ROUGE scores.

🧠 Author

Gustavo Mainchein — AI Solutions Development Specialist - LinkedIn

📜 License

This project is licensed under the MIT License.