MindDiff, git diff for mental models

Company: MindDiff — a self-reported educational tool that uses AI to diagnose student misconceptions in coding assignments by running experiments instead of guessing.

What Changed: The project description states the team built a system that diagnoses false beliefs behind student code errors using GPT-5.6 and Pyodide sandboxed execution, aiming to move beyond traditional autograders or AI tutors that only report failure or provide fixes.

Single Most Important Open Question: Is there evidence of real-world use or traction with students or instructors? The description contains no data on adoption, usage, or feedback from actual users.

The description states that MindDiff is a tool for diagnosing student misconceptions in coding assignments. It reads failing Python submissions, hypothesizes the false belief behind the error using GPT-5.6, and designs discriminating probes to test each hypothesis via execution in a Pyodide sandbox. The system determines whether the misconception is confirmed, ruled out, or unverifiable based on execution results.

Inference: The tool appears to be built for educational use, specifically in programming education, with an emphasis on Socratic questioning and diagnostic feedback over direct correction.

The description states that MindDiff was inspired by the problem of AI-generated homework leading to "quietly disappearing understanding" because current tools (autograders and AI tutors) do not diagnose why a student is wrong, only that they are. The tool aims to prove its diagnosis through execution rather than assertion.

Inference: The positioning is that MindDiff is a diagnostic tool for education that moves beyond simple feedback to deeper understanding by using experimental verification.

The description states that the tool is intended for students and instructors in programming education, particularly those working with Python fundamentals. It mentions an instructor dashboard with heatmaps and clustering of misconceptions, suggesting it targets educators who want to understand class-wide patterns.

Inference: The primary customer segments are likely students learning programming and instructors teaching introductory coding courses.

The description does not state anything about pricing or a business model. It only describes the tool’s functionality and how it was built.

Not evidenced

The project was built using:

• Stack: Next.js, TypeScript, Drizzle ORM over Turso/LibSQL, Zod validated structured outputs from OpenAI SDK, Pyodide for sandboxed Python execution.
• Core Engine: Built in a single Codex CLI session on GPT-5.6 Terra.
• Deployment: Vercel/Turso.

The system includes:

• A Pyodide sandbox for executing probes
• Structured output from GPT-5.6
• Rate limiting and input validation

Inference: The tool is built with modern web stack and uses LLMs for reasoning, with sandboxed execution to avoid server-side code risks.

The description states that the system works end-to-end on real submissions, including edge cases. It mentions stress testing and fixing bugs in prompts and cost exploits, but does not provide any data on:

• Number of users
• Usage frequency
• Feedback from educators or students
• Classroom piloting or integration

Not evidenced

The description does not mention any competitors or market context beyond the general problem of autograders and AI tutors.

Not evidenced

1. No Traction: The tool is described as a hackathon project with no evidence of real-world adoption.
2. Unverified Claims: The system claims to prove diagnoses through execution, but the description does not include validation or testing beyond internal stress tests.
3. Single Developer: The team size is listed as one (Jugger Naut), which may limit scalability or product development speed.
4. No Pricing or Monetization Strategy: No indication of how this would be monetized in a real product.

1. Has the system been tested with actual students and instructors? What feedback have you received?
2. How do you plan to scale beyond a single developer?
3. Are there any plans for classroom piloting or integration with LMS platforms?
4. What is your path to monetization, if any?
5. Can you demonstrate how the system handles edge cases in real-world usage?