Curriculum-Based Bangla Study Assistant for NCTB School Students
An AI-Powered RAG System for Textbook-Based Question Answering and Exam Practice
- Muhsina Tarannum Munfa
- S.M Seefat Alam
- Fariha Ahmed
- Miraj Uddin Chowdhury
- Bangla NLP
- Natural Language Processing
- RAG
- Large Language Model
- NCTB
- Educational AI
- Question Generation
- Question Answering
- MCQ
- Creative Question
- Automated Grading
1. Project Overview
Curriculum-Based Bangla Study Assistant for NCTB School Students is an AI-based study assistant designed for students following the Bangladesh NCTB school curriculum. The system uses NCTB textbook content to answer student questions and provide exam-style practice.
The system supports three main learning modes: Ask Question, MCQ Practice, and Creative Question (CQ) Practice. It retrieves relevant textbook content and uses an LLM to generate answers or questions based on that content.
2. Objectives
- To provide textbook-based answers to students' questions.
- To generate MCQs and Creative Questions based on NCTB textbook content.
- To make generated questions closer to the style and difficulty of real curriculum questions.
- To provide automated grading and feedback for student answers.
- To provide a simple study interface for NCTB school students.
3. System Architecture
The system consists of four main parts:
Textbook Retrieval
NCTB textbooks are cleaned, divided into meaningful chunks, and stored in a vector database. Each chunk contains metadata such as class, subject, chapter, and page. When a student asks a question, relevant textbook passages are retrieved.
Question Pattern Library
A collection of curriculum-style MCQ and CQ examples is used to guide the LLM on question format, wording, and difficulty. These examples are used for question style, not as the main knowledge source.
Generation and Grading
The generation engine receives relevant textbook passages and generates answers for questions asked by students, MCQs, or CQs for practice. It also keeps reference answers internally and grades student responses with feedback.
Frontend
The frontend provides the student interface. Students can select their class, subject, chapter, and practice mode, then interact with the system and view answers, questions, marks, and feedback.
4. Technologies Used
- Python
- FastAPI
- Large Language Model (LLM)
- Retrieval-Augmented Generation (RAG)
- Vector Database (Chroma/FAISS)
- NCTB textbook dataset
- HTML/CSS-based frontend
- JSON-based question pattern examples
5. Main Features
- Textbook-based Question Answering
- MCQ Generation and Practice
- Creative Question Generation
- Difficulty-based Question Generation
- Automated Answer Grading
- Feedback for Student Answers
- Source-grounded responses using textbook passages
- Class, subject, and chapter-based learning
6. Conclusion
This project combines Bangla NLP, RAG, and LLM-based generation to create a curriculum-focused study assistant for NCTB students. Instead of generating completely general educational content, the system uses retrieved textbook passages as the main source of knowledge and uses curriculum examples to guide question style. This makes the system more suitable for textbook-based learning and exam preparation.