E-Tutor Exam City – Contextual RAG Chatbot with Cache

May 2025

An advanced AI-powered tutoring platform with voice support, contextual RAG-based Q&A, and MCQ generation for all grades and university subjects.

Project Demo

Technologies Used

PythonDockerRedisOpenAI GPT-4QdrantAngularText-to-SpeechHeyGenWhisper

Project Links

Live Demo

Project Overview

E-Tutor Exam City is an intelligent tutoring system that leverages a Retrieval-Augmented Generation (RAG) architecture to provide context-aware answers and generate personalized MCQs from a rich academic corpus. The system is trained on a comprehensive dataset spanning school-level textbooks (Grade 1–12), university-level curriculum, and past exam papers. Users can interact via voice or text, receiving responses through both a HeyGen avatar and text-to-speech output. Redis-powered caching enables high-speed delivery of both context and MCQs. The full stack—including an Angular frontend and Python backend—is containerized using Docker, ensuring seamless scalability and deployment.

Key Features

Multi-Level Curriculum Training

Trained on books and papers from Grade 1 to university level including past exams.

MCQ Generation

Dynamically generates MCQs for each grade and subject based on context.

Voice Interaction

Supports microphone input and verbal responses using TTS.

Contextual Q&A

Delivers highly relevant answers using RAG and semantic chunking.

Smart Caching

Caches context and MCQs using Redis for faster response time.

HeyGen Avatar

Presents an interactive AI tutor in the frontend for a realistic experience.

Dockerized Architecture

Fully containerized using Docker for scalable deployment.

Challenges

•Training a generalized RAG model across all grade levels and domain-specific subjects using textbooks and past papers
•Managing dynamic context search across multiple Qdrant databases per subject and grade
•Delivering fast performance with real-time document chunking and Redis caching
•Coordinating multi-modal interaction with HeyGen avatar, text-to-speech, and whisper-based voice input

Key Learnings

•Engineered a smart educational RAG pipeline capable of cross-domain content understanding
•Implemented Redis caching for both MCQs and context passages to improve latency and scale
•Designed and optimized Qdrant database routing logic to handle diverse educational data in a modular way
•Integrated HeyGen avatar in real-time UI pipeline, improving user engagement and delivery realism