AI Code Mentor

An AI-powered programming mentor that explains code, finds bugs, and teaches beginners how to think using Google Gemini.

Comment

Вот тебе полностью готовый Project Story для Devpost — уже в Markdown, с заголовками, жирным, логикой и без воды. Просто скопируй и вставь 👇

AI Code Mentor — Teaching Developers How to Think, Not Copy

## Inspiration

When we started learning programming, we noticed something frustrating: every time we asked ChatGPT for help, it gave us the final answer — but we didn’t actually learn.

We could copy the code. We could submit the assignment. But when we saw a similar problem again, we were lost.

We realized something important: AI is making answers easier, but thinking harder.

So we asked a different question:

What if AI didn’t solve problems for you — but instead trained you to solve them yourself?

That idea became AI Code Mentor.

## What it does

AI Code Mentor is a learning-focused AI that reviews code the way a great teacher would.

Instead of saying:

“Here is the fixed code.”

It says:

“What do you think will happen if this array is empty?” “Why do you think this loop never stops?” “What value does this variable really have right now?”

This creates a Learning Loop:

  1. The student writes code
  2. The AI asks guiding questions
  3. The student thinks
  4. The student improves the code
  5. The AI explains why it is better

We call this Socratic Mode — learning through questions, not copy-paste.

## How we built it

We built the project using Google Gemini 3 inside Google AI Studio.

Gemini 3 is used for:

  • Understanding the structure of the code
  • Detecting logical and runtime errors
  • Generating human-like teaching questions
  • Comparing the old and improved versions of the code

Instead of a simple “fixer” prompt, we designed a multi-step reasoning prompt that forces the model to:

  1. Analyze what the code is trying to do
  2. Find mistakes
  3. Explain them simply
  4. Guide the user with questions
  5. Only then show an improved version

This turns Gemini from an answer machine into a thinking coach.

## Challenges we faced

The hardest part was stopping the AI from giving the answer too early.

Large language models are optimized to be helpful — and “helpful” usually means “give the solution.” We had to carefully design prompts so Gemini would:

  • Ask first
  • Explain second
  • Solve last

Another challenge was designing explanations that are simple enough for beginners but still technically correct.

## What we learned

We learned that:

  • The best learning happens when people struggle a little
  • Questions are more powerful than answers
  • AI can be a teacher, not just a shortcut

We also learned how to use Gemini 3 not just as a chatbot, but as a reasoning engine that guides human thinking.

## Why this matters

As AI becomes more powerful, there is a real risk that people stop thinking and start copying.

AI Code Mentor goes in the opposite direction:

It makes people smarter by forcing them to think.

That’s the future we want.

Built With

Share this project:

Updates