AI Agent with Function Calling (Google Gemini)

Project Overview

This project is a Python-based AI agent that leverages the Google Gemini API to perform tasks by interacting with external tools through function calling. The primary goal of this project was to explore the practical application of Large Language Models (LLMs) in building autonomous agents that can execute real-world actions, such as fetching live data from an API.

The agent is capable of understanding a user's request, determining which available tool is needed to fulfill it, and then calling that tool with the correct parameters.

Key Features & Design

• LLM Integration: Utilizes the Google Gemini API to power the agent's reasoning and decision-making capabilities.
• Function Calling: Implements the function calling pattern native to the Gemini models, allowing the LLM to request the execution of predefined Python functions.
• External Tooling: Includes example tools that can be called by the agent, such as fetching the current price of a stock or the current weather in a given location.
• Structured Output: The agent is designed to receive structured data (JSON) from the tools it calls and use that data to formulate a natural language response.
• Extensible Architecture: The system is designed to be easily extensible, allowing new tools and functions to be added to expand the agent's capabilities.

How It Works

1. A user provides a prompt to the agent (e.g., "What is the current price of NVDA?").
2. The prompt, along with a list of available Python functions, is sent to the Google Gemini API.
3. The Gemini model analyzes the prompt and determines that the get_stock_price function is needed. It then returns a request to call that function with the appropriate argument (ticker="NVDA").
4. The Python script executes the local get_stock_price function, which fetches the data from an external financial API.
5. The data returned by the function is sent back to the Gemini model in a second API call.
6. The model uses this new information to generate a final, natural language answer for the user (e.g., "The current price of NVDA is $123.45.").

How to Run

To run the agent, you will need a Google API key with access to the Gemini API.

1. Install Dependencies:

   pip install google-generativeai

2. Set API Key: Set your Google API key as an environment variable.
   • On Linux/macOS:

     export GOOGLE_API_KEY='your_api_key_here'

   • On Windows:

     set GOOGLE_API_KEY='your_api_key_here'

3. Run the Agent: From the root of the project directory, run the main script:

   python3 src/main.py

---

This project was built as part of the backend development curriculum on Boot.dev to explore advanced AI agent architecture.