Text Recognition Agent Tutorial

A step-by-step tutorial for building an OCR (Optical Character Recognition) AI agent using LangGraph that can extract and process text from images, with progressively added tools for enhanced functionality.

Overview

This project demonstrates how to create an AI agent that can:

• Extract text from images using vision-capable language models
• Preprocess images to improve OCR accuracy
• Convert temperatures between Fahrenheit and Celsius
• Convert measurements (length, weight, volume) between different units

The tutorial is designed to show how agents can grow in complexity by incrementally adding new tools and capabilities.

Project Structure

text-recognition-agent/
├── agent/
│   └── tools.py              # Tools that the agent can access
├── additions/                # Tools to be added incrementally
│   ├── additions_1_opencv.py    # Image preprocessing tools
│   ├── additions_2_temperature.py  # Temperature conversion
│   └── additions_3_unit_conversions.py  # Unit conversion tools
├── images/
│   └── chocolate_cake_recipe.png  # Sample image for testing
├── main.py                   # Main agent implementation
└── README.md

Project set up

The tutorial starts with the agent having access to one tool, which allows the agent to do OCR text extraction from images using vision-capable LLMs (OpenAI GPT-4o or Ollama Qwen2.5-VL).

Tools can then be copied to tools.py and imported to main.py to extend the functionality of the agent.

1. Image Preprocessing (additions_1_opencv.py)
2. Fahrenheit to Celsius conversion (additions_2_temperature.py)
3. Unit Conversions (additions_3_unit_conversions.py)

Requirements

• Python 3.13.5
• Virtual environment with the following packages:
  • numpy
  • opencv-python
  • pyyaml
  • requests
  • langchain-core
  • langchain-openai
  • langchain-ollama
  • langgraph

Usage

Basic OCR

Run the main script to extract text from the sample image.

Adding New Tools

The tutorial aims to demonstrate how you can add incremental complexity by moving tools from the additions/ folder to agent/tools.py:

• Start with basic OCR - Use only extract_text tool
• Add image preprocessing - Import tools from additions_1_opencv.py
• Add temperature conversion - Import from additions_2_temperature.py
• Add unit conversions - Import from additions_3_unit_conversions.py