EduRAG - Multilingual NCERT Doubt Solver using OPEA-Based RAG Pipeline :

🚀 Overview :

EduRAG is a curriculum-locked, multilingual AI tutoring system that answers student doubts strictly from NCERT textbooks (Grades 6–10). Unlike general-purpose LLMs, this system:

• Does NOT use internet knowledge
• Does NOT hallucinate
• Uses only NCERT textbooks
• Provides chapter-level citations
• Supports English, Hindi & Urdu

Built using an OPEA-based modular RAG architecture optimized for Intel CPUs.

🎯 Problem It Solves :

In Indian classrooms:

• Students ask doubts in different languages
• General AI tools hallucinate answers
• No system provides textbook-verified citations
• Most solutions require GPU/cloud infrastructure

EduRAG solves this by:

• Restricting knowledge to NCERT
• Enabling cross-lingual retrieval
• Running entirely on local CPU
• Providing explainable answers with citations

🧠 Architecture (OPEA-Based RAG):

This project follows the Observe → Plan → Execute → Act loop:

I. Observe

• Detect language (EN / HI / UR)
• Normalize subject & grade

II. Plan

Apply metadata hard-filtering:

• Grade
• Subject
• Chapter

III. Execute

• Retrieve relevant chunks using FAISS
• Augment prompt with NCERT context
• Generate answer using Qwen2.5-0.5B-Instruct

IV. Act

• Provide chapter-level citations
• Store session history
• Collect student feedback

🧩 Tech Stack:

1.Backend

• FastAPI
• LangChain
• FAISS
• HuggingFace Transformers
• Qwen2.5-0.5B-Instruct
• Multilingual-E5-Small
• Helsinki-NLP Translation Models

2.Frontend

• HTML5
• CSS3
• Vanilla JavaScript

🌐 Multilingual Support:

Subject Type	Query Language	Output
Hindi	Hindi	Hindi
Urdu	Urdu	Urdu
Science	EN / HI / UR	Same as query
Maths	EN / HI / UR	Same as query
Social Sci	EN / HI / UR	Same as query

Cross-lingual semantic mapping is enabled using multilingual embeddings.

📊 Performance:

• ⏱ Latency: 2.5 – 4.5 seconds (CPU)
• 📚 40+ NCERT textbooks indexed
• 🎯 >85% grounded response accuracy
• 🔎 Top-k metadata-filtered retrieval

📁 Project Structure:

├── backend/
│   ├── main.py
│   ├── chat_memory.py
│   ├── feedback.py
│   └── feedback_store.json
│
├── rag/
│   └── rag_pipeline.py
│
├── vectorstore/
│   └── faiss_index/
│
├── frontend/
│   └── index.html
│
├── data/
│   └── ncert_pdfs/
│
└── README.md

▶️ Running Locally:

• Install dependencies - pip install fastapi uvicorn torch transformers langchain faiss-cpu
• Start backend - uvicorn backend.main:app --reload
• Open frontend - Open frontend/index.html in browser.

🔐 Zero Hallucination Policy:

If answer is not found in NCERT: "This question is outside the NCERT curriculum."

No external knowledge is ever used.

🎓 Academic Context:

Focus Areas:

• Retrieval-Augmented Generation
• Multilingual NLP
• Educational AI
• Explainable AI
• CPU-Optimized Inference