I’m currently double majoring in Physics and Mathematical Sciences at KAIST.
During exam season, I kept seeing friends upload entire lecture PDFs, homework sets, and solution files into GPT-style chat interfaces and ask things like:
- “Explain this.”
- “Summarize this.”
- “Make practice problems.”
- “Predict what will be on the exam.”
That works sometimes, but it felt inefficient.
For actual exam preparation, the hard part is not just summarizing lecture notes. The hard part is tracking the recurring solution patterns in a specific course, the professor’s homework style, the areas that appear repeatedly, and the mistakes I keep making.
So I built PAIDEIA.
GitHub: https://github.com/TaewoooPark/PAIDEIA
What is PAIDEIA?
PAIDEIA is a local-first Claude Code plugin that turns your own course materials into a course-specific exam preparation knowledge base.
You put lecture notes, homework PDFs, solution PDFs, and past exam material into a course folder. PAIDEIA then helps generate a persistent study graph for that course.
It is designed around one idea:
Your course, your patterns, your errors, your cheatsheet.
Instead of treating exam prep as a fresh chat every time, PAIDEIA keeps the artifacts as plain Markdown files inside your course folder.
What it does
PAIDEIA can help with:
- Converting lecture / homework / solution PDFs into Markdown
- Extracting recurring solution patterns from reference solutions
- Mapping homework coverage to estimate high-probability exam areas
- Generating pattern cards
- Building a weakness map from your mistakes
- Creating one-page exam cheatsheets
- Generating twin variants of existing problems
- Generating mock exams
- Reading handwritten answer PDFs through OCR
- Comparing your answer against a reference solution
- Accumulating mistakes into future drills and cheatsheets
Everything is stored as editable Markdown.
No locked-in app state. No opaque dashboard. No “start from zero” every time.
Why I made it
Most AI study tools try to become a friendlier tutor.
But for university STEM courses, I think students often need something slightly different: a course-specific memory system.
Exams are usually tied very closely to a particular professor’s lecture notes, homework distribution, notation, and solution style. Even when the concepts are standard, the way they appear in assignments and exams is often highly local.
So PAIDEIA focuses less on being a chatbot and more on building a study graph from your own materials.
For example:
- If a solution technique appears repeatedly in homework, PAIDEIA can turn it into a pattern card.
- If homework coverage is dense in one topic, PAIDEIA can mark it as a likely exam zone.
- If you make the same mistake while solving problems, PAIDEIA can record it in a weakness map.
- If you generate a cheatsheet later, those mistakes can shape what appears on it.
The goal is not just to answer questions. The goal is to make your preparation compound.
Local-first by design
PAIDEIA is intentionally local-first.
The outputs live in your course folder as plain files. You can open them in any editor, sync them however you want, or use them with tools like Obsidian.
The project is also open source.
Repository: https://github.com/TaewoooPark/PAIDEIA
There is also a Codex CLI edition here:
https://github.com/TaewoooPark/PAIDEIA-codex
Current status
This is still early, and I am testing it mainly on math, physics, and engineering-style courses.
I’m especially interested in feedback from students or developers who:
- Study from lecture PDFs and homework solutions
- Use Claude Code or similar agentic coding tools
- Prefer local-first workflows
- Want editable Markdown artifacts instead of closed study apps
- Have tried using LLMs for exam prep and found the workflow messy
Feedback wanted
If this workflow sounds useful, I’d love feedback on:
- The plugin structure
- The study graph concept
- The OCR / handwritten grading workflow
- The Markdown artifact format
- What an actually useful exam-prep agent should generate
I built this because I wanted exam prep to be less like repeatedly renting a chatbot session, and more like slowly building a permanent map of what I’m learning and where I’m still weak.
Top comments (0)