OranGPT

OranGPT is a Streamlit app for classifying text as formal or informal in English and Polish. It combines a lightweight web UI, pre-trained scikit-learn models, and LIME-based highlighting so users can see both the predicted tone and the words that influenced it most.

Live demo: https://orangcorp-orangpt.streamlit.app/

Features

• Bilingual tone analysis for English and Polish.
• Choice of model from the sidebar, with pre-trained models shipped in the repository.
• Tone summary with confidence and a formality index.
• Word-level highlighting to show which parts of the text pushed the decision toward formal or informal language.
• Session-state driven Streamlit UI so results stay visible while the sidebar changes.

Tech Stack

• Python
• Streamlit
• pandas
• scikit-learn
• LIME
• joblib

Repository Layout

orangpt/
├── app.py                  # Streamlit entry point and UI
├── ai/
│   ├── __init__.py
│   ├── classifier.py       # Model loading, training, and prediction logic
│   └── models/             # Pre-trained model artifacts and metadata
├── assets/
│   └── logo.png            # Application branding asset
├── extracted_data/         # Training data and data prep helpers
├── tests/
│   └── test_classifier.py  # Unit tests for classifier behavior
├── utils/
│   ├── __init__.py
│   └── text_utils.py       # Text normalization and highlighting helpers
├── docs/
│   └── raport.md           # Project report and methodology notes
├── requirements.txt        # Python dependencies
└── README.md

How It Works

The app takes user input, routes it through the selected language and model, then returns:

1. A formal/informal label.
2. A confidence score.
3. A formality probability used to render the visual gauge.
4. Highlighted text that marks influential formal and informal tokens.

The current repository ships a standard classifier pipeline backed by TF-IDF features and an ensemble of classical machine-learning models. Model artifacts for English and Polish are stored in ai/models/.

Prerequisites

• Python 3.10 or newer.
• A virtual environment is recommended.

Local Setup

git clone <repository-url>
cd orangpt
python -m venv venv
source venv/bin/activate
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

Configuration

The app reads optional environment variables from a local .env file:

• ORANGPT_APP_TITLE - custom title shown in the Streamlit header.
• ORANGPT_COMMUNITY - community name used in the subtitle.

Example:

ORANGPT_APP_TITLE=OranGPT Formality Checker
ORANGPT_COMMUNITY=Orang

Run the App

streamlit run app.py

Then open the local Streamlit URL printed in the terminal.

Testing

python -m unittest discover tests

Data and Models

• Training data lives in extracted_data/ as compressed CSV files for English and Polish formality classes.
• Pre-trained model files are stored in ai/models/.
• If a requested standard model file is missing, the classifier code can retrain and save it again.

Deployment

The project is deployed on Streamlit at:

https://orangcorp-orangpt.streamlit.app/

Project Notes

• The UI is built around Streamlit session state, so analysis results persist while users adjust sidebar settings.
• The visualization uses color-coded highlights to distinguish formal and informal signals.
• The project report in docs/raport.md contains the broader methodology and background for the formality-classification work.