Sentence-Level-AI-text-detection

Sentence-Level AI Text Detection

A full-stack application to detect AI-generated content in text at the sentence level, leveraging LSTM, DistilBERT, and BERT models. The backend is built with FastAPI, and the frontend is a React application.

Table of Contents

• Features

• Tech Stack

• Repository Structure

• Prerequisites

• Installation

  • Backend Setup
  • Frontend Setup

• Running the Application

  • Start Backend
  • Start Frontend

• API Endpoints

• Usage

• Configuration

• Contributing

• License

Features

• Sentence-level AI vs. human detection
• Combined scoring using BERT probability and edit distance
• Switchable models: LSTM and DistilBERT for sentence classification
• FastAPI backend with CORS support
• React frontend with live word count and colored confidence slider

Tech Stack

Layer	Technology
Backend	Python, FastAPI, TensorFlow, PyTorch, Transformers, nltk
Frontend	React, Axios, CSS
Models	LSTM, DistilBERT, BERT
Package Mgmt	pip, npm

Repository Structure

Sentence-Level-AI-text-detection/
├── backend/
│   ├── main.py
│   ├── requirements.txt
│   ├── ai_detection_model_lstm.h5
│   └── distilbert_ai_detector/
├── frontend/
│   ├── package.json
│   ├── src/
│   │   ├── App.js
│   │   ├── App.css
│   │   └── ...
│   └── public/
│       └── index.html
└── README.md

Prerequisites

• Python 3.8+
• Node.js 14+
• Git
• (Optional) CUDA-enabled GPU for faster model inference

Installation

Backend Setup

1. Clone the repository:

   git clone https://github.com/Gourabh09/Sentence-Level-AI-text-detection.git
   cd Sentence-Level-AI-text-detection/backend

2. Create a virtual environment:

   python3 -m venv venv
   source venv/bin/activate   # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install --upgrade pip
   pip install -r requirements.txt

4. Download or train models:

   • Place ai_detection_model_lstm.h5 and tokenizer_lstm.pkl in the backend folder.
   • Place your DistilBERT model directory named distilbert_ai_detector here.

Frontend Setup

1. Navigate to frontend:

   cd ../frontend

2. Install dependencies:

   npm install

Running the Application

Start Backend

From the backend/ directory:

uvicorn main:app --reload --host 0.0.0.0 --port 8000

Start Frontend

From the frontend/ directory:

npm start

The React app will open at http://localhost:3000 and proxy API calls to the FastAPI backend at http://localhost:8000.

API Endpoints

Method	Endpoint	Payload	Description
GET	/	—	Health check
POST	/predict	`{ text: string, model_choice: "lstm"	"distilbert" }`	Returns AI score and sentence-level predictions

Sample Request

curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d '{"text": "Your input text here.", "model_choice": "distilbert"}'

Usage

1. Enter or paste your text in the textarea.

2. Select the detection model (LSTM or DistilBERT).

3. Click Upload to see:

   • AI Detection Score (combined BERT + edit distance)
   • Sentence-Level Predictions with confidence bars.

Configuration

• MAX_LENGTH (512) and DISTILBERT_MAX_LENGTH (256) can be adjusted in main.py.
• Model paths and CSV data paths are defined at the top of main.py.

Contributing

Contributions are welcome! Please fork the repo and open a pull request with your improvements.