Document Analysis Graph: Educational Example with LangGraph and Tools

This project demonstrates how to use LangGraph with tool integration for document and image analysis, using Google Gemini models via LangChain. It is designed for educational purposes and showcases how to build a simple agent that can:

• Extract text from images using a vision-capable LLM
• Perform basic computations (e.g., division)
• Route tasks between an assistant and tool nodes in a graph structure

Features

• Vision Tool: Extracts text from images (e.g., meal plans, notes)
• Computation Tool: Performs division operations
• Graph-based Agent: Uses LangGraph to manage control flow between assistant and tools
• Notebook-based: All code and examples are in a Jupyter notebook for easy experimentation

Example Usage

1. Extract Text from an Image

   • Place an image (e.g., Batman_training_and_meals.png) in the project directory.
   • Run the notebook cell to extract text from the image using the extract_text tool.

2. Ask Questions or Perform Calculations

   • Example: "Divide 6790 by 5" or "What should I buy for the dinner menu according to the note?"
   • The agent will use the appropriate tool and return the answer.

Getting Started

1. Clone the repository and install dependencies:

   pip install -r requirements.txt

2. Set up your Google API key (and Langfuse keys if using tracing):
   • Create a .env file with:

     GOOGLE_API_KEY=your_google_api_key
     LANGFUSE_PUBLIC_KEY=your_langfuse_public_key
     LANGFUSE_SECRET_KEY=your_langfuse_secret_key
     

3. Open doc-analysis-graph.ipynb in Jupyter and follow the cells.

Educational Notes

• The project is structured to show how to combine LLMs, tool use, and graph-based control flow.
• The agent ("Alfred") is designed to be extendable with more tools or logic.
• The code is commented for clarity and learning.

References

• LangGraph Documentation
• LangChain Google GenAI
• Google Gemini API
• Hugging Face Agents Course – educational material

---

For educational use only.