AI Agent from Scratch

This project is a simple yet powerful implementation of an AI agent built from scratch using Hugging Face's InferenceClient and integrating basic tools like a calculator. The agent is capable of interacting with users, performing arithmetic operations, and responding intelligently based on the context.

Features

• AI Agent: A custom AI agent built with Hugging Face's models that can interact with users, interpret queries, and provide responses.
• Tool Integration: A built-in calculator tool to perform arithmetic operations like addition, subtraction, multiplication, division, and more.
• Verbose Logging: Logs are maintained for every agent interaction, providing insights into internal processes such as tool usage and thoughts.
• Streamlit Interface: A web-based user interface built with Streamlit for interacting with the agent and viewing verbose logs.

Installation

Prerequisites

• Python 3.11 or higher
• Poetry for managing dependencies and environments (or use pip if you prefer)

Step-by-step Installation

1. Clone the Repository

   git clone <repository-url>
   cd ai-agent-from-scratch

2. Install Dependencies using Poetry

   If you don't have Poetry installed, follow the instructions on Poetry's website to install it.

   Then, install the project dependencies:

   poetry install

3. Setup Environment Variables

   Create a .env file in the root directory and add your Hugging Face API token as follows:

   hf_api=YOUR_HUGGING_FACE_API_TOKEN

4. Run the Application

   Once the dependencies are installed and the environment is set up, you can run the Streamlit app:

   streamlit run app.py

   This will start the application locally, and you can access it via your browser at http://localhost:8501.

How It Works

1. Agent Initialization: The agent is initialized with the Hugging Face API token and available tools like the calculator. The model used for the agent is microsoft/Phi-3.5-mini-instruct.
2. User Interaction: The user types a query into the input field, and the agent decides whether to provide a direct response or use a tool.
3. Tool Usage: If the query requires an external tool, like the calculator, the agent generates the appropriate tool action and invokes it.
4. Verbose Logs: Every interaction with the agent, including tool usage and thought processes, is logged and displayed in the right column of the UI for transparency.

Example Queries

• Basic Chat: "How are you?"
• Math Operations: "What is 125 times 165?"
• Geometric Calculations: "What is the area of a square with side length of 16.25?"

Code Overview

app.py

This file contains the Streamlit frontend that handles the UI and user interaction with the agent. It provides an interface for both the user and the verbose logs.

agent.py

This file contains the Agent class that handles the logic of interacting with the Hugging Face API, using tools, and maintaining the chat history.

tools.py

This file defines the available tools for the agent, like the calculator. The calculator performs basic arithmetic operations based on user input.

prompts.py

Defines the prompts used to guide the agent in various scenarios. It includes the system prompt, tool instructions, and general guidelines.

test.py

A script to test the functionality of the agent with sample queries. Useful for validating the agent's responses and tool usage.

pyproject.toml

Configuration for managing dependencies, versioning, and project setup using Poetry.

Dependencies

The project relies on the following Python libraries:

• huggingface-hub: For interacting with Hugging Face models.
• ipykernel: For running Jupyter kernels.
• python-dotenv: For managing environment variables.
• streamlit: For building the web interface.

Troubleshooting

If you encounter any issues:

• Ensure that your Hugging Face API token is valid.
• Check if the environment variables are correctly set in the .env file.
• Make sure all dependencies are installed using poetry install or pip install.

License

This project is open-source and available under the MIT License.