Physics Chatbot: Offline AI-Powered Learning Assistant

Project Overview

The Physics Chatbot is an offline AI-driven educational tool designed to assist students of grades 9-10 in learning physics. It provides accurate and textbook-aligned answers, conceptual clarity, and interactive explanations based on the Physics Classes 9-10 (English Version) by NCTB. By leveraging Retrieval Augmented Generation (RAG) and lightweight AI models, this chatbot bridges the digital divide and ensures equitable access to quality education.

Features

• Offline Functionality: Works without internet, reducing reliance on cloud servers.
• Textbook Alignment: Retrieves relevant content directly from the NCTB textbook.
• Interactive Explanations: Provides clear and concise explanations with inline LaTeX for mathematical equations.
• Efficient Memory Usage: Stores conversation summaries using Small Language Models (SLMs) to enhance context retention.
• Sustainability: Optimized for minimal resource consumption, ensuring energy efficiency.

Technologies Used

• AI Framework: Retrieval Augmented Generation (RAG)
• Model Execution: Ollama for LLM/SLM
• GPU-Accelerated Computation: Google Colab
• Database: MongoDB for storing chat history
• Frontend: Inline LaTeX and Markdown for equations
• Programming Languages: Python

System Architecture

1. Planning Phase:

   • Implemented a simple RAG system with chunked textbook data stored in a vector database.
   • Retrieved relevant chunks and utilized prompt engineering with LLM for response generation.

2. Advanced RAG System:

   • Generated related questions for enhanced retrieval.
   • Summarized responses using SLM to reduce memory overhead and improve precision.

3. Frontend Development:

   • Integrated inline LaTeX for mathematical clarity.
   • Designed a user-friendly interface.

Dataset

The dataset includes factual, conceptual, explanatory, and math-based problems sourced from:

• Physics Classes 9-10 (NCTB)
• Panjeree Test Paper
• Teachingbd24.com

Evaluation Metrics

• BERT Score
• BLEU
• ROUGE
• Cosine Similarity
• Online Judge Evaluations

Manual Testing Feedback

Feedback from diverse testers, including students and educators, highlighted the following:

• Strengths:

  • Clarity of explanations (4.5/5)
  • Helpfulness in understanding concepts (4.4/5)
  • Appropriate language and tone

• Challenges:

  • Struggled with out-of-scope questions.
  • Lacked mathematical notation in some responses.
  • Slightly slow response time.

• Suggestions:

  • Improve response conciseness.
  • Add features like graph generation and enhanced mathematical notations.

Installation and Setup

1. Clone the repository:

   git clone [https://github.com/your-username/physics-chatbot.git](https://github.com/tan7vir/Physics-Chatbot-using-Advanced-RAG)

2. Navigate to the project directory:

   cd physics-chatbot

3. Install dependencies:

   pip install -r requirements.txt

4. Run the backend:

   cd backend
   python run app.py

5. Run the Frontend:

   cd ..
   cd Frontend
   streamlit run app.py

Future Improvements

• Fine-tuning AI models specifically on the NCTB Physics textbook.
• Incorporating support for multiple sessions and context continuity.
• Enhancing retrieval mechanisms for better accuracy.
• Reducing response time through optimization.
• Integrating graph generation and improved mathematical outputs.

Contributors

• Md Tanvirul Islam Niloy - Lead Developer & Database
• Abu Saleh Al Nahian - Database

---

We welcome contributions! If you’d like to improve this project, please submit a pull request or open an issue.