Virtual Tutor

An intelligent adaptive learning platform that uses Bayesian Knowledge Tracing to personalize educational content and track student mastery across multiple subjects.

Project Overview

Virtual Tutor was developed as a final year project for the Bachelor of Computer Science program at Nanyang Technological University (completed April 2020). The platform demonstrates the application of probabilistic machine learning to education, creating a personalized tutoring experience that adapts to each student's knowledge level.

Author: Fad Rahim Institution: Nanyang Technological University Year: 2020

Key Features

Bayesian Knowledge Tracing

• Adaptive Difficulty: Questions automatically adjust based on estimated student mastery
• Real-time Assessment: Continuous evaluation of knowledge level using Bayesian inference
• Personalized Learning Paths: Content difficulty scales with demonstrated competency
• Statistical Modeling: Uses PyMC3 for probabilistic programming and knowledge estimation

Multi-Subject Support

Comprehensive content coverage across:

• Mathematics
• Physics
• Chemistry
• Biology

Learning Modes

• Study Notes: Review core concepts and theory
• Practice Sessions: Reinforce learning with adaptive exercises
• Assessments: Test knowledge with intelligent quizzes
• Progress Tracking: Monitor improvement over time

How Bayesian Knowledge Tracing Works

The platform uses a sophisticated probabilistic model to estimate student knowledge:

1. Prior Belief: Starts with a uniform Beta distribution representing initial uncertainty
2. Evidence Collection: Observes student performance on quiz questions
3. Bayesian Updating: Uses Bernoulli likelihood to update knowledge estimates
4. Logistic Transformation: Maps latent knowledge to observable success probability
5. MAP Estimation: Calculates Maximum A Posteriori estimate of mastery level

The algorithm adapts question difficulty when students demonstrate >70% mastery, ensuring optimal challenge and engagement.

Technology Stack

• Backend: Flask (Python web framework)
• Database: SQLite with Flask-SQLAlchemy ORM
• Machine Learning: PyMC3, Theano for Bayesian inference
• Frontend: HTML5, CSS3 (Bootstrap), JavaScript (jQuery)
• Forms: Flask-WTF with WTForms validation

Installation

Prerequisites

• Python 3.7+
• pip package manager

Setup Instructions

1. Clone the repository

   git clone https://github.com/yourusername/Virtual-Tutor.git
   cd Virtual-Tutor

2. Create virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\\Scripts\\activate

3. Install dependencies

   pip install -r requirements.txt

4. Run the application

   python run.py

5. Access the platform Open your browser and navigate to:

   http://localhost:4000
   

Usage

1. Select a Subject: Choose from Mathematics, Physics, Chemistry, or Biology
2. Pick a Topic: Select specific units or concepts to study
3. Choose Learning Mode:
   • Study notes for review
   • Practice for skill building
   • Assessment for testing
4. Take Adaptive Quizzes: Answer questions and receive real-time knowledge estimates
5. Progress Through Difficulty Levels: The system automatically advances you when ready

Project Structure

Virtual-Tutor/
│
├── model/                    # Bayesian Knowledge Tracing logic
│   ├── __init__.py
│   └── calc.py              # BKT calculation engine
│
├── virtualtutor/            # Main application package
│   ├── __init__.py          # Flask app factory
│   ├── routes.py            # URL routes and views
│   ├── models.py            # Database models
│   ├── forms.py             # WTForms definitions
│   │
│   ├── static/              # CSS, JavaScript, images
│   │   ├── css/
│   │   └── js/              # Quiz logic, adaptive difficulty
│   │
│   └── templates/           # Jinja2 HTML templates
│       ├── layout.html      # Base template
│       ├── home.html
│       └── ...
│
├── run.py                   # Application entry point
├── requirements.txt         # Python dependencies
├── .gitignore
└── README.md

Algorithm Details

Bayesian Model Specification

with pm.Model() as model_bkt:
    # Prior: Beta(1,1) - uniform distribution
    theta = pm.Beta('Beta', alpha=1, beta=1)

    # Logistic transformation
    p = pm.Deterministic('p', 1.0 / (1 + exp(-(theta * 10 - 5))))

    # Likelihood: Bernoulli outcomes
    x = pm.Bernoulli('x', p=p, observed=student_data)

# Estimate knowledge level
map_estimate = pm.find_MAP(model=model_bkt)
knowledge_level = map_estimate['Beta'] * 100

Features in Detail

Adaptive Question Selection

• Initial questions assess baseline knowledge
• Success triggers progression to harder content
• Struggles result in review of fundamentals
• Maintains optimal challenge through Bayesian updates

Performance Tracking

• Real-time mastery percentage display
• Historical performance data
• Subject-specific progress analytics

Future Enhancements

Potential improvements for the platform:

• Integration with modern ML frameworks (TensorFlow Probability, PyTorch)
• Expanded question banks with community contributions
• Detailed analytics dashboards
• Mobile responsive design
• Multi-language support
• Collaborative learning features

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Nanyang Technological University, School of Computer Science and Engineering
• PyMC3 development team for probabilistic programming tools
• Flask community for excellent web framework documentation

Contact

Fad Rahim For questions or collaboration opportunities, feel free to reach out!

---

This project demonstrates the practical application of Bayesian inference to create adaptive educational technology. It showcases skills in full-stack development, machine learning, probabilistic modeling, and software engineering.