PDF Parser Agent

Overview

This project provides an AI-powered solution for parsing complex PDF files, specifically designed to extract structured data such as Haircut Schedules from CME-related documents. It leverages the pydantic-ai framework, Google Gemini models, and PyPDF2 to read, process, and extract information from multi-page, table-rich PDFs. The extracted data is returned in a structured JSON format, making it easy to integrate with other systems or workflows.

Features

• Parses multi-page, complex PDFs with tables and sections.
• Focuses on extracting Haircut Schedule data (asset class, duration, value, etc.).
• Uses AI agents for iterative, high-accuracy extraction.
• REST API for easy integration and automation.

How to Run

1. Install Dependencies

Make sure you have Python 3.8+ installed. Then, install the required packages:

pip install -r requirements.txt

2. Set Your Gemini API Key

Export your Gemini API key as an environment variable before running the agent or API:

export GEMINI_API_KEY={your_api_key_here}

3. Run the AI Agent Directly

You can run the agent directly to parse a PDF file and print the extracted data:

python agents/pydantic_ai_agent.py

• By default, it expects a file named acceptable-collateral-futures-options-select-forwards.pdf in the working directory. You can modify the filename in the script as needed.

4. Run the REST API

Start the FastAPI server to use the agent via a REST API:

python -m uvicorn api.rest_api:app --reload

• The API will be available at http://localhost:8000.
• Upload a PDF via the /upload/ endpoint using a tool like curl or Postman:

curl -X POST "http://localhost:8000/upload/" -F "file=@path_to_your_pdf.pdf"

4. Project Structure

• agents/pydantic_ai_agent.py: Main AI agent logic for PDF parsing.
• api/rest_api.py: FastAPI REST API for PDF upload and parsing.
• knowledge/: Example files for testing (PDF, CSV, TXT).
• requirements.txt: Python dependencies.

Notes

• The agent is optimized for CME Haircut Schedule PDFs but can be adapted for similar structured documents.
• For best results, use high-quality, text-based PDFs (not scanned images).

License

This project is for educational and research purposes. See individual file headers for more details.