🧠 AI Writing Detector

This project uses a hybrid classifier that combines linguistic features with RoBERTa embeddings to detect AI-generated text.

🔧 Model Weights Required

To run the prediction API, you must manually download the following files:

File/Folder Name	Description
hybrid_classifier_weights.pt	PyTorch weights for the hybrid model
feature_extractor.pkl	Pre-trained linguistic feature extractor
tokenizer/	HuggingFace tokenizer folder (roberta-base)

📁 Where to Place the Files

Place the files in the root directory of the project like this:

server/ 
├── app.py 
├── hybrid_classifier_weights.pt 
├── feature_extractor.pkl 
├── tokenizer/ 
│ ├── tokenizer_config.json
│ ├── vocab.json 
│ └── merges.txt 
├── model_classifier/

⚠️ These files are not included in the GitHub repository due to their size. Please obtain them from the Google Drive.

🚀 Getting Started

Backend

1. Install dependencies

pip install -r requirements.txt

2. Run the backend

cd server
python app.py

Frontend

1. Install dependencies

npm install

2. Run the frontend

cd ai_writing_detector
npm run dev

✅ Example

1. Paste a paragraph of text with at least 50 words into the text box
2. Click "Check"
3. The model will return:

Prediction: Human-written or AI-generated
Confidence Score: A visual confidence score of AI generated from 0–100%