AI home tutor

The problem

Standard LLMs often act as answer engines rather than teachers. They give away solutions too early, fail to track a student's recurring mistakes, and lack the pedagogical structure needed for actual learning and retention.

What it does

• Enforces a three-phase teaching loop that prioritizes diagnosis and prerequisite checks before providing any explanations.
• Maintains a persistent state file in your workspace to track user error history and ensure past mistakes are reviewed in future sessions.
• Implements "Knowledge-Base-First" retrieval to prevent hallucinations by prioritizing your local study materials and textbooks over pre-trained data.
• Uses T-Shape explanation logic that leads with formal definitions before moving to misconception analysis and cross-disciplinary analogies.
• Dynamically adjusts difficulty based on real-time grading of practice problems.

Frameworks & tools

Works with local markdown files, PDF study materials, and knowledge bases like Notion or Obsidian via RAG. Supports LaTeX for mathematical notation.

Why this beats prompting it yourself

This skill isn't a simple "act as a tutor" prompt. It includes a mandatory state-management system to track learning progress across sessions and a rigid multi-turn logic that prevents the model from "leaking" answers during the diagnostic phase.

Use cases

• Preparing for IB, AP, or A-Level exams using specific curriculum textbooks.
• Reviewing complex STEM concepts where visual analysis of graphs or diagrams is required.
• Automated grading of homework with structured "Highlights" and "Needs Improvement" feedback.
• Long-term mastery of a subject through spaced repetition of past errors.

Known limitations

Visual analysis requires the user to manually upload images for the agent to process. Does not support non-educational tasks like general software engineering or weather updates.