Tiger Mom Project

Introduction

Tiger Mom is an AI agent prototype that uses LLM to monitor user activity during their work.

🚀 If you want to see a video demo: https://youtu.be/LbUZTz5JUIM

Motivation

The goal is to offer a one-click solution to avoid distractions. Unlike browser-based extensions such as StayFocusd, this AI agent is far more flexible as it analyzes the actual content on the user's screen rather than blocking entire websites.

How It Works

1. The prototype takes screenshots of the user's device.
2. Calls OpenAI's GPT-4o model to analyze screenshot contents.
3. Calls OpenAI's GPT-4o-mini model to compare the screenshot content with the user's defined focus task and distraction settings.
4. Why GPT-4o for Vision? It is cheaper to inference via the API than GPT-4o-mini for vision tasks, despite GPT-4o-mini being a lighter model.

GUI Implementation

• Built using PyQt5 instead of web frameworks.
• A desktop application is necessary as it requires local system privileges to take screenshots.
• Future improvements include fine-tuning and deploying a local model to replace GPT-4o for screenshot vision analysis, as screenshots may contain private information.

Try Our Prototype Out Yourself

Install Dependencies

After cloning the project onto your local device, install the required dependencies:

pip install -r requirements.txt

It is recommended that you create a new development environment before doing so.

Set Up OpenAI API Key

Since Vision LLM is expensive, costing around $0.002-0.003 per screenshot analysis, you need to set up your own OpenAI API key as an environment variable. Go to OpenAI API to create your own API key. Then, navigate to your project directory via terminal and use the commands below:

On macOS/Linux:

export OPENAI_API_KEY="your-api-key-here"

On Windows (Command Prompt):

set OPENAI_API_KEY="your-api-key-here"

On Windows (PowerShell):

$env:OPENAI_API_KEY="your-api-key-here"

Run the Prototype

Now you are ready to run the prototype! Navigate to the src folder and run:

python main.py

Future Steps

• Reduce LLM vision inference cost
• Enhance GUI design for a better user experience
• Containerize the application for easier access