🐱‍👤 Detext

Detext is a web-based text classification tool that determines whether a given text is "safe" or "fraud". It uses technologies such as React, FastAPI, Pydantic, Huggingface's Transformers to offer a robust solution for text classification tasks. The Detext's dataset is built from the three collated fraud/scam/spam datasets.

---

🎯 Purpose

To classify text as either "safe" or "fraud" using machine learning models.

---

🚀 Features

• 🧠 Classifies text to determin if it's safe or fraud
• 💻 Frontend built with React for a seamless user experience
• ⚡ Fast and responsive backend with FastAPI
• 📜 API documentation available at /docs for easy interaction
• 🧪 Easy to train and customize model

---

🛠️ Tech Stack

• Frontend: React, SASS
• Backend: FastAPI, Pydantic
• Modeling: Huggingface Transformers, DistilBERT (uncased)

---

📦 Development

# Clone the repo
git clone https://github.com/markcalendario/detext.git
cd detext

Client

# Navigate to the client directory
cd client

# Install dependencies
npm install

# Start the development server
npm run dev

Server

# Navigate to the server directory
cd server

# Install dependencies
pip install -r requirements.txt

# Start the FastAPI server
fastapi dev main.py

Trainer

To train the model, follow these steps:

Create a virtual environment:

cd trainer
python -m venv venv

Activate the virtual environment:

.\venv\Scripts\activate

Create a Jupyter kernel for the virtual environment:

pip install ipykernel

python -m ipykernel install --user --name=detext-venv

# Restart the code editor to switch to the new kernel

Run the training notebook:

Open the notebook trainer.ipynb and train the model using Huggingface transformers.

---

🤖 After Training

Once the model is trained, copy the model files to the backend to be used for predictions.

# Copy the trained model to the server's model dir
cp -r trainer/out/detext_dataset/* server/model/detext/

Ensure the model files include:

• The model checkpoint (e.g., pytorch_model.bin or similar)
• The model configuration (e.g., config.json)
• Tokenizer files (e.g., tokenizer.json)

---

📄 Usage

Once the project is set up and running, you can:

1. Run the React frontend:

   • Open a web browser and navigate to http://localhost:5173 (default port for React).
   • Input text into the provided form, and the system will classify whether it is safe or fraud.

2. Access FastAPI Documentation:

   • The FastAPI backend provides interactive API documentation at http://localhost:8000/docs.
   • Here you can view available endpoints, try them out, and see example requests and responses for the text classification service.

🚀 Deployment

# Build and run client and server concurrently using docker compose
docker compose up --build -d