AI-Detector (AI Detection Hub)

A simple Streamlit app that predicts whether input text is AI‑generated or human‑written using a scikit‑learn model. The app preprocesses text (cleanup, stopword removal, stemming) and runs inference with a joblib‑serialized model.

Features

• Streamlit UI with a text area and one‑click prediction
• Text preprocessing: regex cleanup, tokenization, stopword removal, stemming (NLTK)
• scikit‑learn model loaded from models/model.pkl
• Lightweight and easy to run locally or in Codespaces

Project structure

app.py                 # Streamlit app
requirements.txt       # Python dependencies
models/
	model.pkl            # Trained classification model (joblib)
dataset/
	archive.zip          # Dataset artifact (optional, for training)
notebooks/
	AI_Detector.ipynb    # Exploration/training notebook (optional)

Requirements

• Python 3.9+
• Internet access on first run (to download NLTK resources like punkt and stopwords)

Setup

Create a virtual environment and install dependencies:

python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -r requirements.txt

If you’re using VS Code, the workspace is configured to pick .venv automatically.

Optional: pre-download NLTK data to avoid runtime prompts:

python - << 'PY'
import nltk
nltk.download('punkt')
nltk.download('stopwords')
PY

Run

source .venv/bin/activate
streamlit run app.py

If you’re on a remote environment (e.g., Codespaces) and need relaxed CORS/XSRF for previews:

streamlit run app.py --server.enableCORS false --server.enableXsrfProtection false

Use

1. Paste or type text into the text area.
2. Click “Run Detection”.
3. See the predicted label and a simple confidence indicator.

Notes:

• The model is probabilistic; treat results as guidance, not ground truth.
• The app performs basic preprocessing before inference.

Model

• Path: models/model.pkl
• Format: joblib (scikit‑learn pipeline or estimator)
• Training: See notebooks/AI_Detector.ipynb (if provided) for exploration/training; the app itself does not train.

Troubleshooting

• FileNotFoundError for model

  • Ensure models/model.pkl exists and the path in app.py matches.
  • If your model has a different filename, update the load path in app.py accordingly.

• NLTK resource not found (e.g., punkt or stopwords)

  • Ensure internet access on first run so nltk.download(...) can fetch data, or pre-download using the snippet above.

• Version mismatch loading the model

  • Use the dependencies from requirements.txt. If the pickle was created with a very different scikit‑learn version, re‑export the model with this environment or retrain.

• Streamlit doesn’t open a browser automatically

  • Open the URL shown in the terminal (default http://localhost:8501). In remote environments, use the forwarded port.

License

No license file is included. Add one if you plan to distribute.