PrepMate: WASSCE Exam Preparation Assistant

PrepMate is an AI-powered chatbot designed to help students prepare for the West African Senior School Certificate Examination (WASSCE). It leverages a domain-specific dataset and a transformer-based language model to provide accurate, helpful answers to exam-related questions in Math, English, Science, and more.

🚀 Live Demo

Try the app here: https://prepmate.streamlit.app/

---

Features

• Domain-specific question answering for WASSCE subjects
• Similarity matching with existing exam questions
• Confidence scoring for answers (progress bar and badges)
• User feedback mechanism (thumbs up/down, suggestions)
• Example/sample questions for user guidance
• Warning and fallback for out-of-domain queries
• Clean, modern Streamlit interface with icons and loading spinners

Project Structure

prepmate/
├── app.py                # Streamlit web interface
├── fine_tune_gpt2.py     # Model training script (if using local model)
├── requirements.txt      # Project dependencies
├── data/                 # Dataset directory
│   └── waec_qa_dataset.jsonl
├── models/               # Saved model checkpoints (gitignored)
├── README.md             # Project documentation
└── ...

Dataset

• Format: JSONL, each line is a JSON object with at least a question and answer field.
• Example:

{"question": "What is the chemical formula for water?", "answer": "H2O"}

• Used for similarity matching, sample questions, and (optionally) model fine-tuning.

How It Works

1. User submits a question
2. Similarity check: The app compares the question to those in the dataset
3. If similar: The model generates an answer and displays confidence
4. If not similar: The app warns the user and shows sample/example questions
5. Feedback: Users can rate answers and provide suggestions

Running Locally

1. Clone the repo and cd prepmate
2. Create a virtual environment and activate it
3. Install dependencies:

   pip install -r requirements.txt

4. (Optional) Download NLTK data:

   python -c "import nltk; nltk.download('punkt')"

5. Run the app:

   streamlit run app.py

Model

• Uses a transformer-based model (e.g., GPT-2) via Hugging Face Transformers
• Can be extended to use OpenAI API for more powerful responses
• Model is guided by the domain-specific dataset for relevance

Evaluation

• BLEU score and perplexity (if fine-tuning locally)
• Qualitative: User feedback, confidence scores, and similarity checks