Explain My Mistake

An adaptive AI tutor powered by Gemini 3 that turns mistakes into personalized learning moments.

Comment

Inspiration

Most learning tools focus on giving the right answer.
But in real classrooms, the most common student frustration is:

“I don’t understand **why* my answer is wrong.”*

This gap becomes even more visible for younger learners, multilingual students, and learners who rely on voice or accessibility-first experiences.

The inspiration for Explain My Mistake came from observing how human teachers work: they diagnose misunderstandings, adapt explanations, and ask the right next question. I wanted to explore whether Gemini 3 could enable that same kind of human-centered, adaptive tutoring experience.

🧠 What I Learned

Through this project, I learned that:

  • Effective AI tutoring is less about answers and more about reasoning and diagnosis
  • Adaptive follow-up questioning creates much deeper understanding than static explanations
  • Accessibility, safety, and age-awareness are not add-ons — they must be designed from the start
  • Gemini 3 excels when used as a thinking partner, not just a text generator

I also learned how to design meaningful AI behavior within the constraints of AI Studio, using prompt architecture instead of traditional UI controls.

How I Built the Project

The project was built using Google AI Studio with the Gemini 3 Pro model.

Core workflow:

  1. The learner provides the Problem Context and How They Tried
  2. Gemini 3 analyzes the reasoning process to detect the underlying misconception
  3. The tutor explains the mistake, highlights a Key Insight, and adapts its teaching style
  4. Gemini 3 generates targeted practice questions based on the detected misconception
  5. A Learning Snapshot summarizes progress and understanding

Key Gemini 3 capabilities used:

  • Deep reasoning over incorrect logic
  • Automatic subject domain detection (math, science, writing, general learning)
  • Adaptive follow-up question generation
  • Mode-aware teaching (Kids Mode vs Standard Mode)
  • Multilingual and audio-first interaction support
  • Safety and parental consent enforcement

For example, if a learner confuses multiplication with addition, Gemini 3 identifies the misconception and explains: [ 7 \times 8 = 8 + 8 + 8 + 8 + 8 + 8 + 8 ] instead of simply stating the correct answer.

🚧 Challenges I Faced

One of the biggest challenges was designing rich interaction without custom UI controls. AI Studio does not allow buttons, checkboxes, or real-time state tracking, so features like:

  • parental consent gating
  • session resets
  • practice answering
  • time tracking

had to be implemented through clear conversational logic and prompt design.

Another challenge was preventing the model from defaulting to “coding mode” when users typed short, plain-language inputs—especially on mobile. This was solved by adding automatic segment detection with a safe fallback to “General” learning.

Balancing simplicity for kids with depth for older learners also required careful prompt tuning to avoid over-simplification or cognitive overload.

🌍 Impact & Vision

Explain My Mistake demonstrates how Gemini 3 can power:

  • mistake-driven learning
  • adaptive pedagogy
  • inclusive, accessible education
  • responsible AI design for children

The project shows that AI tutoring can move beyond answers toward understanding, helping learners turn “oops” moments into real “aha!” breakthroughs.

Built with ❤️ using Gemini 3 via Google AI Studio.

Built With

Share this project:

Updates