ThinkTrace AI — Don't Just Mark the Wrong Answer, Diagnose the Thinking Behind It

Описание: Most classroom tools tell a teacher THAT 40% of the class got a question wrong.
None of them tell the teacher WHY — and the why is the only part that changes
what you do next.

ThinkTrace AI reads each student's written reasoning, groups the whole class by
HOW it is confused rather than by which option was ticked, and then walks each
individual student through a repair path that does not close until they can
explain the concept in their own words without the original misconception in it.

This is a full end-to-end run in 3 minutes: teacher creates a live classroom,
student joins with a code, the question arrives in real time, the AI builds the
class confusion map, and the student works through diagnosis → prerequisite
repair → targeted practice → alternative explanations → teach-back verification.

Watch what happens at the end. The student types "i need more time to
understand" and the system REFUSES to mark the misconception resolved. That is
the point of the whole thing: nothing is marked fixed until the student can
actually explain it.

CHAPTERS
0:00 Why marking the answer isn't enough
0:14 The teacher creates a live classroom
0:31 A student joins with the code
0:55 The question arrives in real time
1:25 LecturePulse — the class confusion map
1:44 ConceptLens — diagnosis and prerequisite repair
2:01 ErrorTwin — practice against the reasoning pattern
2:12 Explain My Way, PerspectiveLab, and Telugu
2:37 Teach-back — and what the teacher sees

WHAT IT DOES
• LecturePulse — groups the class by reasoning, not by answer choice. Surfaces
ranked missing prerequisites, a two-minute intervention, a counterexample for
the board, and a follow-up question that separates the groups.
• ConceptLens — names the belief behind the wrong answer, explains why that
reasoning fails, and gives a counterexample with a confidence score.
• Prerequisite Detective — the shortest repair path through the concept graph.
• ErrorTwin — fresh practice aimed at the reasoning pattern, in contexts the
student has never seen.
• Explain My Way — the same idea as simple language, technical, step by step,
visual, or a real-world analogy, in English, Telugu, Kannada, Hindi, Spanish
or French, with technical terms optionally kept in English.
• PerspectiveLab — the concept re-explained by people whose work depends on
getting it right.
• Teach-back verification — explain it in your own words, and the AI checks
whether the original misconception is actually gone.

BUILT WITH
Next.js 15 (App Router) · React 19 · TypeScript · Tailwind CSS v4 ·
Supabase (Auth, Postgres, Realtime) · Row-level security on all 10 tables ·
Claude Opus 5 with forced tool-use for structured JSON output · Vercel

Runs with zero API keys — clone it and every feature works against a
deterministic analyzer, with every screen stating which backend produced what
you are reading.

LINKS
Live app: https://aha-loop-orcin.vercel.app
Source: https://github.com/ParishruthiGanesh/AhaLoop

#AI #EdTech #MachineLearning #Claude #NextJS #Supabase #Hackathon #LearningScience