Open Education : AI-Powered Course Generation Platform

Overview

This project is an AI-powered course creation and delivery platform designed to streamline how educational content is generated, reviewed, and consumed. It supports a structured workflow with three primary roles:

• Admin

• Tutor (Faculty)

• Student

The system combines role-based access control, automated content generation, and a custom-built media pipeline to produce full-length educational videos from simple textual inputs.

Stack: Next.js, Node.js, NestJS, MinIO S3, PostgreSQL, Redis, BullMQ

Authentication & User Flow

To ensure a smooth and minimal-friction experience, authentication is handled exclusively via Google OAuth.

Once a user logs in:

• The backend determines whether the user is new or returning

• New users are routed to onboarding

• Existing users are redirected to their respective dashboards

Onboarding System

Tutor (Faculty) Onboarding

Tutors go through a verification-based onboarding process:

• Pay an initial fee of ₹1000

• Submit:

  • Educational credentials

  • Bank account details

  • Additional metadata

After submission:

• Admin reviews the application

• Tutors can be approved or rejected

Only approved tutors can create and publish courses.

Student Onboarding

Students complete a lightweight onboarding process:

• Provide basic personal and academic information

• No verification is required

Core Features

Student

• Browse and enroll in courses

• Track learning progress

Tutor

• Create courses using natural language input

• Define difficulty level and constraints

• Review and edit AI-generated course structures

• Trigger full course generation

AI-Assisted Course Generation

Tutors provide a simple textual description of a course. The system processes this input to generate:

• A structured course roadmap

• Chronologically ordered topics and subtopics

Each subtopic is treated as an independent unit and results in a separate video.

Media Generation Pipeline

The platform uses a fully custom-built pipeline to generate videos without relying on traditional video generation systems.

1. Course Breakdown

Each course is structured as:

Main Topic → Subtopics

Each subtopic corresponds to one video.

2. Contextual Input Handling

For each subtopic, the system uses:

• Main topic

• Current subtopic

• Compressed context from previous subtopics (key concepts summary)

This ensures continuity while avoiding redundant data processing.

3. Script & Slide Generation

• A structured README file is generated in slide format

• Content is organized for presentation-ready output

4. Transcript Generation

• A detailed transcript is created

• Includes timestamped segments for precise synchronization

5. Audio Generation

• Transcript is split based on timestamps

• Audio is generated for each segment independently

6. Slide Rendering

• Slides are converted into images

• Each image corresponds to a specific segment

7. Incremental Video Assembly

• Video is built segment by segment

• Images and audio clips are aligned using timestamps

• Segments are appended progressively

This prevents full regeneration in case of failure.

8. Storage & Access Control

• Videos are stored securely

• Access is restricted via backend-generated temporary links

• Content is served only through the platform domain

Job Processing & Reliability

The system uses a queue-based architecture for handling heavy background tasks.

• Each subtopic video is processed as an independent job

• Jobs are split into multiple stages:

  • Content generation

  • Audio generation

  • Image generation

  • Video compilation

Reliability Features

• Maximum of 3 retries per stage

• Redis-based progress tracking

• Restartable stages with persistent intermediate outputs

State Management

Each video follows a defined state machine:

CREATED → SCRIPT_READY → AUDIO_READY → VIDEO_READY → COMPLETED

This provides:

• Clear observability

• Easier debugging

• Controlled retry mechanisms

System Architecture

• Fully containerized using Docker

• Backend and services orchestrated locally

• Designed for asynchronous processing using queues

Performance Constraints

• System currently runs on local infrastructure

• Handles approximately 10 concurrent course generations

• Beyond this, performance degradation occurs due to resource-heavy processing

What Makes This System Unique

• End-to-end automation from text input to video output

• Custom-built media generation pipeline

• Fine-grained control over each stage of content creation

• Strong fault tolerance with resumable workflows

Links

• Backend Repository: https://github.com/aniketduttaAD/open-education-backend

• Frontend Repository: https://github.com/aniketduttaAD/open-education-frontend

Summary

This project demonstrates how AI can be integrated into a structured system to automate complex educational workflows. Instead of relying on monolithic tools, the platform focuses on modularity, control, and reliability, making it adaptable for future scaling and enhancements.