Project: Create a Personal AI Study Assistant

In this project you will design a personal AI study assistant — a custom-prompted tutor that explains, quizzes, and tracks what you have learned across any subject. The deliverable is a single system prompt plus a small set of "lesson templates" you can reuse for every new topic you want to learn.

1. Introduction

A free chat window is the worst classroom you can imagine. There is no syllabus, no quizzes, no spaced repetition, no record of what you covered yesterday. Most people lose 70% of what an AI tutor explained simply because they never had to retrieve it again.

This project fixes that. We will build a study assistant that turns any chat session into a structured learning loop: explain, check understanding, quiz, summarise, schedule a review. The whole system runs on prompts — no app required.

2. The Concept Explained

A study assistant is just a chat with a strong system prompt, plus a discipline around how each session ends. The system prompt sets the persona and rules. The session structure makes sure the learning sticks.

The clever part is the notes log. It is just a Markdown file or a Google Doc. Each lesson ends with the assistant generating a five-line summary, a small quiz, and a recommended review date. You paste that into your notes log. Next time you start a session you paste the previous summary back in as context. The "memory" of the system is your file, not the model.

3. The Problem Without a Structured Assistant

Here is what most people type when they want to learn something with AI:

explain probability to me, I'm a beginner

4. The Solution: A System Prompt + Lesson Loop

You are my personal study assistant. Your job is to help me learn,
not to dump information.

Rules:
- Calibrate to my current level. Ask 1 short question before explaining
  anything new, to gauge what I already know.
- Explain concepts using simple language and one concrete analogy.
- Never give me more than 4 short paragraphs in a single message.
- After each explanation, ask one "check-understanding" question and
  wait for my answer before continuing.
- At the end of a lesson, generate:
   (a) a 5-line summary I can paste into my notes
   (b) a 3-question quiz (mix recall + application)
   (c) a "next review" date using spaced repetition
       (1 day, 3 days, 7 days, 21 days)
- If you don't know something, say so. Do not invent facts.
- Track open questions I raised but we didn't fully resolve.

Subject: {paste your subject here, e.g. "introductory probability"}
My current level: {beginner / intermediate / advanced}
My goal: {paste your concrete goal, e.g. "pass the data science
mid-term in 4 weeks"}

Last lesson summary (paste from notes):
"""
Topic: conditional probability basics.
Covered: P(A|B), tree diagrams, simple Bayes' rule example.
Quiz score last time: 2/3.
Open questions: when is independence safe to assume?
"""

Today I want to learn: Bayes' theorem in plain language and one
worked example involving medical testing.

Begin with a check-in question. Don't dump theory at me.

Notice what just happened. The assistant did not lecture. It asked a calibration question. That single design choice — forced by the system prompt — is what separates studying from scrolling.

End the lesson now. Output exactly this structure:

### Summary (5 lines, no fluff)
- ...

### Quiz (3 questions)
1. (recall)
2. (recall)
3. (application — give me a tiny scenario, not a definition)

### Open questions to revisit
- ...

### Next review
Suggested date: {today + N days} where N follows my spaced-repetition
schedule. Justify the choice in one line.

5. Step-by-Step Breakdown

1. Set the persona once. Save the system prompt as a snippet, a saved prompt, or a custom GPT. You should not be rewriting it each session.
2. Always start with the last summary. Pasting the previous summary back in gives the assistant the "memory" it does not natively have. Two paragraphs of context is enough.
3. Resist long answers. The 4-paragraph cap in the system prompt is the most underrated rule. Long answers feel productive but are forgotten faster.
4. Take the quiz seriously. Write your answers down before reading the model's evaluation. The retrieval is the learning — not the explanation.
5. Update your notes log. Paste the summary, quiz score, and next-review date into a single notes file. After a week you will have a real study record.
6. Tune the spaced-repetition schedule. The 1-3-7-21 day schedule is a sensible default. Shorten it for hard subjects, lengthen it once a topic is solid.

6. Practice Exercises

7. Key Takeaways