Excel Worker AI Agent (LangGraph with tool calling)

This Agent enabling analysis of Excel files using free-form queries as user input.

If you have any questions or would like to collaborate, feel free to reach out to me on LinkedIn. You're more than welcome!

Core Architecture

• State-based agent using StateGraph pattern for workflow control
• OpenAI GPT-4o model for query generation and validation
• Two-node system: query generation and validation nodes
• Type-safe state management with AgentState TypedDict

Main Tools

• load_preview_data: Excel structure analysis and metadata extraction
• complex_duckdb_query: SQL operations with DuckDB connection
• simple_dataframe_query: Pandas operations with safe eval environment

Key Features

• State Management:

  • Tracks query execution, validation states
  • Manages conversation history
  • Thread-safe checkpointing
  • Error propagation and recovery

• Tool Orchestration:

  • Dynamic tool selection by LLM
  • Parallel tool execution capability
  • Safety limits: max 3 calls per tool
  • Automatic tool argument injection
  • Tool result validation

• Data Processing:

  • Excel file structure analysis
  • SQL query execution via DuckDB
  • Pandas operations in a safe environment
  • Null/NaN value handling
  • Data type preservation

• Control Flow:

  • Two-stage validation pipeline
  • Max 7 iterations per query
  • Conditional edge routing
  • State-based decision making

Intallation

Prerequisites

• Access to JupyterLab, Google Colab, or another interactive computing environment to run this Jupyter Notebook.
• Access to LLM API.

Step 1: Clone the Repository

Clone this repository to your local machine:

git clone <REPOSITORY_URL>
cd <PROJECT_FOLDER>

Step 2: Open Jupyter Notebook in JupyterLab

Ensure that <PROJECT_FOLDER> is accessible in JupyterLab by setting it as your working directory in JupyterLab.

• In JupyterLab, use the "Open from Path" option to load ExcelWorkerLLMToolCallAgent.ipynb.
• Similarly, load .env and populate the variable keys with appropriate values.
• The first cell in the Notebook installs the required libraries: %pip install langchain langgraph pandas python-dotenv duckdb

Step 3: Run the Jupyter Notebook

To execute the notebook, select each cell and press Shift + Enter.