Catch the AI

• 

  Catch the AI: AI hides 3 mistakes in your notes. Catch and fix them to prove you understood.

Inspiration

I'm a B.Com student from Karnataka. I study at home with only my syllabus copy, no textbooks and no classes. I kept re-reading my notes, but I never knew if I had really understood them or just memorized them. I wanted a way to test myself using my own notes and make it fun.

What it does

You paste your notes and click Learn to get a short summary. When you're ready, the AI secretly plants 3 mistakes in your notes. You tap the sentences you think are wrong and fix them in your own words. The AI checks the meaning, not exact words.

• If you tap a correct sentence, it tells you to look again, so you can't just guess.
• Stuck? Use a hint, or Show answers.
• At the end, the Review screen shows your score, each mistake with the correct fact, and exactly what to revise.

How we built it

I'm not a coder, so I built it with Claude Code using the Devpost skills, step by step: scope, PRD, spec, then building and testing in small slices. - Plain HTML, CSS and JavaScript

• One Vercel serverless function calls the Groq API, so my API key stays hidden and never reaches the browser
• The AI returns structured JSON, and the code double-checks every case before showing it ## Challenges we ran into The hardest part was making the AI's mistakes fair. Sometimes it planted a mistake that wasn't actually wrong in real life, changed two facts at once, or put two mistakes on the same idea. I tested it with my real B.Com notes (Cost Accounting, Research Methodology, Forex), found each unfair mistake myself, and added one rule at a time until the cases became fair. ## Accomplishments that we're proud of I built a full working app without knowing how to code. I'm proud that the mistakes are now fair and that checking works by meaning, so students can answer in their own words.

What we learned

I learned how to give AI clear instructions and test them on examples where I already know the right answer. I also learned that rules make good AI behavior likely, not guaranteed, so the code should always double-check the AI's output.

What's next for Catch the AI

Support for Kannada notes, more subjects, saving progress over time, and a mode for teachers to create cases for their class

Built With

• claude-code
• css
• groq
• html5
• javascript
• vercel