PDF Quiz Lab

cd apps/experiments/pdf-quiz-lab
pnpm install
pnpm dev

PDF Quiz Lab turns source material into a scored multiple-choice quiz. Paste text, load the Vercel Blob sample paragraph, or upload .txt / .md / .pdf. The Route Handler extracts PDF text with unpdf when needed, truncates input to ~12k characters, and calls generateObject for exactly five questions. The client UI lets you answer and score locally - no fabricated quizzes when AI Gateway / OpenAI keys are missing (503).

FeaturesCopy Anchor Link

• Paste or upload – .txt, .md, and .pdf via multipart/form-data, or JSON { text }.
• PDF extraction – unpdf extractText; parse failures return 400 with a clear message (no silent empty quizzes).
• Five MCQs – Zod schema enforces four choices and correctIndex 0–3 plus an explanation.
• Interactive scoring – client-side scoreQuiz from logic.ts after you submit answers.
• Sample text – one-click paragraph about Vercel Blob for a quick demo without a file.
• Honest 503 – never invents questions when no provider is configured.

Server ReferenceCopy Anchor Link

POST /api/quizCopy Anchor Link

Accepts JSON { "text": "..." } or multipart/form-data with text and optional file.

Success (200)Copy Anchor Link

{
  "questions": [
    {
      "question": "What does put() do in Vercel Blob?",
      "choices": ["Upload", "Delete", "List", "Redirect"],
      "correctIndex": 0,
      "explanation": "put() uploads an object."
    }
  ],
  "charsUsed": 312,
  "provider": "ai-gateway"
}

Exactly five questions are required by the Zod schema.

ErrorsCopy Anchor Link

Implementation DetailsCopy Anchor Link

if (!aiProviderAvailable()) {
  return Response.json(AI_SETUP_BODY, { status: 503 });
}

const result = await generateObject({
  model: resolveModel(),
  schema: quizSchema, // questions.length === 5
  prompt: `Create exactly 5 multiple-choice quiz questions...`,
});

Use CasesCopy Anchor Link

• Turn docs or blog drafts into quick comprehension checks.
• Teach generateObject with a fixed-length array schema.
• Show PDF text extraction before sending content to a model.

LimitationsCopy Anchor Link

• Image-only PDFs often fail extraction - paste text instead.
• No auth or rate limits; public deploys can spend your model quota.
• Quiz quality depends on the model and how clear the source text is.
• Source text is capped at ~12k characters.

Use in your projectCopy Anchor Link

Copy app/api/quiz/route.ts, lib/services.ts, lib/ai-model.ts, and the helpers in logic.ts. Keep the 503 response. Wire a form that posts FormData the same way as demo.tsx.

DeploymentCopy Anchor Link

Deploy on Vercel

Prompts for AI_GATEWAY_API_KEY. No Marketplace store is required.

Local DevelopmentCopy Anchor Link

cd apps/experiments/pdf-quiz-lab
pnpm install
# Add AI_GATEWAY_API_KEY (or OPENAI_API_KEY) to .env.local
pnpm dev

curl -s -X POST http://localhost:3010/api/quiz \
  -H "content-type: application/json" \
  -d '{"text":"Vercel Blob is object storage for the web."}'

ConfigurationCopy Anchor Link

Vercel / Next.js Features UsedCopy Anchor Link

• AI SDK generateObject
• Vercel AI Gateway
• Route Handlers (maxDuration = 60)
• unpdf for PDF text extraction
• Zod schemas

Next StepsCopy Anchor Link