I Built an AI Learning Platform as a Student — Here's Exactly How

I built Eduquic because I was tired of passive learning.

I wanted a platform that could generate study content for me, quiz me on it, and track my progress. So instead of waiting for it to exist, I built it.

The Stack (And Why I Chose It)

The Hardest Part: OpenAI Integration

I expected AI to be plug-and-play. It was not.

My first prompt returned content that was too long, weirdly formatted, and sometimes just wrong. I spent a week just on prompt engineering.

What worked:

You are a curriculum designer. Given a topic, return exactly 5 learning modules
in JSON format: { title, summary, keyPoints: string[] }.
Do not add extra text. Return only valid JSON.

Strict output format + strict instructions = consistent results.

The Auth Rabbit Hole

I tried to build auth myself first. Big mistake.

After two days of JWT refresh token logic, I switched to Clerk. Setup took 20 minutes. It handled sessions, OAuth, and UI components out of the box.

Lesson: Use the right tool. Don't reinvent wheels that don't need reinventing.

What I'd Do Differently

• Add rate limiting on the OpenAI calls from day one
• Use Redis for caching generated content (it's expensive to regenerate)
• Build the mobile view first, not last

Eduquic isn't perfect. But it's real, it works, and I built it.

Sometimes that's enough. 🎓