DUX

Agentic course generation with adaptive assessments

ReactFastAPILangGraphGPT-4o

shipped

Introduction

An agentic course generation platform that creates personalized learning experiences using LLM-powered agents, dynamic schema-driven UI, and adaptive assessments.

The platform uses a pipeline of specialized LangGraph agents — each responsible for a stage of course creation — orchestrated through a FastAPI backend and rendered via a schema-driven React frontend.

Pipeline

User Preferences — Learners choose a topic, difficulty level, course length, and learning style through a dynamic form.

Syllabus Architect — A LangGraph agent designs the curriculum — scope, modules, learning objectives, and prerequisite chains.

Lesson Writer — A multi-step agent pipeline researches, drafts, adds interactive checks, self-reviews, and finalizes each lesson.

Quiz & Tutor — Bloom's-aligned assessments are generated per lesson. An adaptive tutor agent analyzes errors and prescribes remediation.

React Frontend  -->  FastAPI Backend  -->  LangGraph Agents  -->  LLM (GPT-4o / Claude)
     |                      |                      |
  Zustand Store        SSE Streaming       Syllabus Architect
  Dynamic UI           REST API            Course Planner
  React Router         Pydantic            Lesson Writer
  Tailwind CSS         MCP Servers         Quiz Generator
                                           Adaptive Tutor

Features

Six core capabilities powering the platform.

AI Course Generation

Enter any topic and the platform generates a full syllabus, lesson content, quizzes, and interactive exercises — all personalized to your level.

Dynamic Schema-Driven UI

Forms, quizzes, and interactive checks are rendered from JSON schemas — supporting text, number, slider, choice, and boolean field types.

Interactive Quizzes

Bloom's taxonomy-aligned assessments with multiple question types, automated grading, and detailed explanations for each answer.

Adaptive Tutoring

An AI tutor agent analyzes quiz performance, identifies error patterns, and provides targeted feedback with specific remediation steps.

Real-time Streaming

Server-Sent Events stream agent progress to the frontend in real time — see each stage of course generation as it happens.

Multi-LLM Support

Swap between OpenAI and Anthropic models via environment config. The LangChain abstraction keeps agent logic provider-agnostic.

Demo

Four screens walk through the full course generation workflow — from initial preferences to adaptive quiz feedback.

Course Setup — Dynamic form captures topic, level, length, and learning style

Live Generation — Real-time SSE stream shows agent progress as lessons are built

Lesson View — Rich markdown with numbered sections, code blocks, and inline checks

Quiz & Feedback — Bloom's-aligned questions with instant grading and explanations

Conclusion

This project demonstrates full-stack AI engineering — from LLM orchestration and agentic pipelines to dynamic, schema-driven frontend rendering.

A modern architecture combining React and Python with LLM-powered agentic workflows for intelligent content generation.

Tech Stack

Frontend

React 19TypeScriptViteTailwind CSSZustandRadix UIReact RouterReact Markdown

Backend

Python 3.11FastAPILangGraphLangChainPydanticUvicornSSE StreamingMCP Servers

AI / LLM

OpenAI GPT-4oAnthropic ClaudeLangChain AgentsStructured OutputBloom's TaxonomyAdaptive Feedback

Infrastructure

DockerDocker ComposeVercelRailway / RenderREST APIJSON Persistence