← Back to work
AI · Learning · 2026

ReTrace

An AI reasoning debugger that finds the first wrong turn in a student's solution, identifies the likely misconception, and turns it into targeted learning material.

Product design · AI workflows · Frontend · Image fits the viewport by default.

Operating context

The prototype connects solution diagnosis with assessments, flashcards, practice, and saved learning material so the feedback can become a next action rather than a final verdict.

Problem

Most AI learning tools reach for an answer before locating the misconception that caused a student to make a mistake.

Approach

The product reviews a typed or photographed solution, identifies the earliest invalid step, tests the likely misconception, then creates a focused learning activity.

Key product decisions

  • Made the first invalid step the primary unit of feedback.
  • Kept diagnosis editable so a learner can correct an input before continuing.
  • Connected explanations to practice and saved material instead of treating them as one-off chat output.

Technical architecture

A Next.js interface translates structured Gemini responses into editable reasoning traces and local learning material.

Current result

A hackathon-ready local prototype with diagnosis, assessments, flashcards, practice, and saved material.

ReTrace interface