AI Text Detection API

An AI/ML API that predicts whether text is human-written or AI-generated. The API offers multiple classification models (Logistic Regression, Random Forest, Neural Networks) paired with different text encoding methods (GloVe word embeddings or TF-IDF vectorization). All of the models are trained on a Kaggle dataset of human and AI-generated essays.

🌐 Live API: http://ec2-3-131-93-133.us-east-2.compute.amazonaws.com/api/v1/

✨ Features

• Classification Models:

• Logistic Regression

• Random Forest

• Neural Network

• Text Encoding:

• GloVe (Global Vectors for Word Representation)

• TF-IDF (Term Frequency-Inverse Document Frequency)

🛠️ Technology Stack

• Programming Language:

  • Python

• Web Framework:

  • FastAPI (REST API)

• Machine Learning:

  • PyTorch (Neural Networks)
  • scikit-learn (Traditional ML)

• Data Processing:

  • Pandas (Data manipulation)

• Deployment:

  • AWS EC2 (Cloud hosting)

• Testing & Documentation:

  • Postman (API testing)

📋 API Documentation

🔌 Endpoints

1. Classification Endpoint

POST /api/v1/classify

Request Body:

{
"text": "Text to classify",
"encoder": "glove",  // Options: "glove", "tfidf"
"model": "logistic_regression"  // Options: "logistic_regression", "random_forest", "neural_network"
}

Response:

{
"model": "Logistic Regression",
"encoder": "glove",
"prediction": "Human",
"confidence": {
    "human": "85.50%",
    "ai": "14.50%"
  }
}

2. Info Endpoint

GET /api/v1/info

Returns model information and usage examples.

⚡ Getting Started

📋 Prerequisites

• Python 3.12+
• Pip package manager

💻 Installation

1. Clone the repository

git clone https://github.com/yourusername/ai-detect.git
cd ai-detect

2. Create and activate virtual environment (optional but recommended)

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Download and prepare the dataset

• Visit Kaggle AI Vs Human Text Dataset
• Download and place the CSV file in ~/Downloads

5. Train the models

• Open src/main.py
• Uncomment the main() function call at the bottom of the file
• Run the training script:

python -m src.main

• This will train all models and save them in src/models/saved/

6. Start the API server

• Open src/api/api.py
• Uncomment the uvicorn line at the bottom of the file
• Run the server:

python -m src.main

The API will be available at http://localhost:8000

📁 Project Structure

ai-detect/
├── src/
│   ├── api/           # FastAPI application
│   ├── models/        # ML model implementations
│   ├── encoders/      # Text encoding modules
│   └── data/          # Data processing utilities
├── tests/             # Unit tests
└── requirements.txt   # Project dependencies

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

👤 Author

Jimmy Andrews

• Email: jcandrews2@icloud.com

📚 Acknowledgments

• Trained using the Kaggle dataset AI Vs Human Text