AI-powered educational tool that diagnoses why students get things wrong — not just that they did.

Every wrong answer traces back to a specific cognitive error. This tool names it. Instead of "Wrong — the answer is X," students receive a named misconception, an explanation anchored to their own words, and the correct understanding in plain language. Teachers get a live class-wide map of which mental models need correcting.

misconception-detector/
├── frontend/          # React + Vite app
│   ├── src/
│   │   ├── components/    # Reusable UI components
│   │   ├── pages/         # Student and Teacher views
│   │   ├── hooks/         # Custom React hooks
│   │   ├── lib/           # API client, utilities
│   │   └── types/         # TypeScript interfaces
│   └── ...
├── backend/           # Express + Node.js API
│   ├── src/
│   │   ├── routes/        # API route handlers
│   │   ├── middleware/     # Auth, logging, error handling
│   │   └── lib/           # Anthropic client, data store
│   └── ...
└── package.json       # Root monorepo scripts

• Node.js 18+
• An Anthropic API key

# Clone and install all dependencies
git clone <repo>
cd misconception-detector
npm run install:all

Backend — create backend/.env:

ANTHROPIC_API_KEY=sk-ant-...
PORT=3001
NODE_ENV=development

Frontend — create frontend/.env:

VITE_API_URL=http://localhost:3001

# Runs both frontend (port 5173) and backend (port 3001)
npm run dev

npm run build    # Builds frontend to frontend/dist/
npm start        # Starts backend (serves API on port 3001)

1. Student picks a topic, reads a question, explains their reasoning
2. Backend forwards the answer to Claude with a carefully engineered system prompt
3. Claude returns a structured JSON diagnosis: misconception name, severity, explanation, correction
4. Student sees the named misconception and where their thinking diverged
5. Teacher view aggregates all submissions into a class-wide misconception frequency map

• Persistence: swap the in-memory store (backend/src/lib/store.ts) for a database (Postgres, Supabase, MongoDB)
• Class codes: add a join-code system so multiple students submit to the same session
• Custom questions: let teachers upload their own questions via the teacher UI
• LMS integration: expose a webhook endpoint compatible with Google Classroom or Moodle

MIT