Agent-as-Coder Challenge

This project contains a coding agent that automatically generates, tests, and self-fixes a Python parser for bank statement PDFs using the Google Gemini API.

Agent Architecture Diagram

The agent operates in a simple, robust loop:

5-Step Run Instructions

1. Clone the Repository

   git clone https://github.com/itz-Mayank/AI_Agent_Data_Parser.git
   cd AI_Agent_Data_Parser

2. Install Dependencies Make sure you have Java installed. Then, install the required Python packages.

   pip install -r requirements.txt

3. Set Your API Key Create a .env file in the root directory and add your Google Gemini API key:

   GOOGLE_API_KEY="YOUR_API_KEY_HERE"
   

4. Add Sample Data Place your bank statement PDF in data/icici/icic_sample.pdf.

5. Run the Agent Execute the agent from your terminal, specifying the target bank.

   python agent.py --target icici

   The agent will begin the process of writing, testing, and fixing the parser, which will be saved in the custom_parsers/ directory.

---

Project Structure

/
├── agent.py                 # The main AI agent script
├── tests/
│   └── test_parser.py       # Validation script to execute the generated parser
├── data/
│   └── icici/
│       └── icici_sample.pdf   # Input PDF for a target bank
├── custom_parsers/
│   └── (Generated by the agent)
├── Output/
│   └── (Generated by the parser)
├── .env                     # For storing your API key
└── README.md

---

Model Information

This agent leverages the power and speed of Google's Gemini family of models. The primary model used during development was gemini-2.5-flash latest.