← 목록

I built a tool that tells me if readers actually understood my articles

devto 2026-08-15 원문 보기 ↗


After publishing 66 technical articles on DevOps and AI infrastructure, I realised I had a blind spot.

I knew how many people visited my articles. I had no idea how many people understood them.

Page views, bounce rates, time on page — none of these tell you whether your explanation of Kubernetes state management actually made sense to the reader.

So I built QuizOps.

What it does

You paste your article URL. GPT-4o reads it and generates 10 multiple-choice questions streamed in real time. You review, edit, and publish. Your readers get a quiz link.

You see who passed, who failed, and which specific questions tripped people up.

The technical bits

The generation uses OpenAI's streaming API with NDJSON output — each question is a complete JSON object on its own line. The frontend reads the ReadableStream line by line and renders questions as they arrive with a fade-in animation. It genuinely looks like the AI is thinking in real time.


// Each line from the stream is a question
const lines = buffer.split('\n');
for (const line of lines) {
if (!line.trim()) continue;
try {
const q = JSON.parse(line);
setQuestions(prev => [...prev, q]);
} catch {}
}

Stack:

Content moderation

Before generating questions, the API runs two classification checks:

  1. Domain intent check — is this URL likely to contain inappropriate content?
  2. Content check — is the fetched article appropriate for educational quiz generation?

Both use a lightweight GPT-4o-mini call before the main generation request.

Community quiz banks

Beyond publisher quizzes, there are community quiz banks (JSON files in the repo) covering: Prompt Engineering, LLM APIs, AI Agents, RAG & Vectors, Kubernetes, Terraform, GitHub Actions, Python for AI Engineers, Cloud Native, and AI Security.

Anyone can add a new bank by dropping a JSON file — no code changes needed.

Try it

quiz.autoshiftops.com — free, no credit card.

Would love feedback from other technical writers on what's missing.