AI Personal Tutor + Self-Learning

What's in this pack

This is the stack for an adult who decided to learn something hard on their own — a new programming language, conversational Spanish, calculus they avoided in college, the AWS Solutions Architect exam. You are the student and the teacher, and the bottleneck isn't access to information (it's never been easier). The bottleneck is structure, active engagement, and memory — turning passive reading into something that actually sticks six months later.

This pack does not assume you have a course or a class. It assumes you have a topic, a chunk of free hours per week, and the willingness to be honest with an AI about what you don't understand. Every pick is open-source or has a clear free tier, and each one is in here because it solves a specific failure mode of solo learning — drifting off-topic, fooling yourself that you understood, forgetting in two weeks what you learned today.

Different from the Teacher / Educator pack: no LMS, no quiz form builder, no slide tool. This is for the learner, not the instructor.

Install in this order (pick topic → tutor prompt → flashcard → quiz → review)

1. Mr. Ranedeer — Personalized AI Tutor Prompt — start here. It's a single GPT-4 / Claude system prompt that turns any chat AI into a tutor that adapts to your level, language, and pace. Zero install. Paste it into Claude or ChatGPT and tell it I want to learn linear algebra; I'm comfortable with Python; treat me like a smart adult who never took the class. You've just hired a 24/7 tutor. Spend a week here before adding anything else.
2. DeepTutor — AI-Powered Personalized Learning Assistant — Mr. Ranedeer hallucinates because it doesn't know your textbook. DeepTutor grounds answers in a document you upload — the PDF of the chapter, the API docs, the practice exam. Now when you ask "what does this equation mean," the tutor cites your book instead of inventing a similar one.
3. Claude Code Agent: Microsoft Study Mode — a Claude Code skill that teaches through guided discovery: it refuses to give you the answer. It asks you what you think the answer is, then nudges. Even outside Microsoft topics, the pattern is the antidote to "AI wrote the code, I didn't learn anything."
4. Claude Code Agent: Demonstrate Understanding — the active-recall agent. Instead of you reading and feeling like you understood, this skill makes you explain back. It asks you to teach the concept in your own words, then catches your hand-waves. Pair with #3: discovery agent for new material, demonstrate-understanding agent before you move on.
5. Claude Howto — Guided Claude Code Learning Path — a structured curriculum (not just a doc dump) for learning Claude Code and the broader agent toolchain. Use it as the template for any "how do I become competent at X" path — module structure, exercises, progress markers. If you're learning coding through AI, start the actual learning here.
6. Claude Reflect — Self-Learning for Claude Code — a reflection skill: at the end of a study session, it walks you through what you learned, what's still fuzzy, what to attack next. This is the closest open-source equivalent to a study journal that asks the right questions back. Run it nightly. Cheap habit, compounds for years.
7. Fabric — AI Prompt Patterns for Everything — the 100+ prompt-pattern library. The patterns you'll live in as a learner: extract_wisdom (textbook chapter → clean outline), summarize_paper, create_quiz (turn a chapter into 15 questions for self-test), analyze_claims (when something in the material smells wrong). One CLI, consistent shape.
8. notebooklm-py — NotebookLM CLI + Python API Skill — NotebookLM's killer feature for learners is audio overviews: upload 4 PDFs from your reading list, get a 12-minute podcast-style discussion you can listen to on a commute. The CLI lets you script that across an entire syllabus instead of clicking through the web UI.
9. Khoj — Your AI Second Brain — semantic search over your own notes, PDFs, and chat history. After 6 weeks of study you'll have 200 markdown notes, 30 PDFs, 50 chat transcripts with your tutor. Khoj is how you ask "didn't I cover Bayes' rule somewhere" and actually find it — without forcing you into the rigid Anki review schedule (use Khoj-as-retrieval instead of spaced-repetition flashcards, or alongside them).
10. Mem0 — Memory Layer for AI Applications — the piece that makes #1, #2, #3, #4 actually personalized. Without persistent memory, every tutor session restarts: "what's your level? what do you already know?" With Mem0, the tutor remembers you got stuck on closures last Tuesday and opens with that on Thursday. The single biggest unlock for solo learning.

How they fit together — the daily / weekly loop

  Pick topic
      │
      ▼
  Mr. Ranedeer  ◄────── Mem0 (remembers your level)
      │
      ▼  asks: what do you know?
  DeepTutor (your textbook PDF) ──► answers grounded in source
      │
      ▼
  Microsoft Study Mode ──► guided discovery (you do the thinking)
      │
      ▼
  Demonstrate Understanding ──► you teach it back, agent catches gaps
      │
      ▼
  Fabric: create_quiz on chapter ──► 15 self-test questions
      │
      ▼
  Claude Reflect (end of session) ──► what's solid? what's fuzzy?
      │
      ▼
  Khoj (next week) ──► search your notes when something feels familiar

The four-tool spine is Ranedeer + DeepTutor + Demonstrate Understanding + Mem0: prompt to talk to, source to ground, active recall to check, memory to remember. Everything else is amplification. NotebookLM is for when you have 4 PDFs and a commute. Khoj is for month two when your notes get unwieldy.

Tradeoffs you'll hit

• Mr. Ranedeer vs DeepTutor — Ranedeer is a prompt (zero infrastructure, works in any LLM, hallucinates freely). DeepTutor is a platform (you deploy or use the hosted version, requires you to actually upload your textbook, but cites it). Use Ranedeer for early exploration of a topic; switch to DeepTutor once you have specific source material.
• Khoj vs Anki (spaced repetition) — Anki is the legendary flashcard tool with the empirically-validated review schedule. It's not in this pack because Anki is not AI-native — making good cards is the entire skill, and AI hasn't fully cracked that. Khoj is the AI-native alternative: instead of forcing yourself to review 50 cards/day, you search semantically when curious. Both work. Anki is more rigorous; Khoj is more humane. If you've tried Anki before and bounced, try Khoj.
• Microsoft Study Mode (Socratic) vs just asking ChatGPT — asking ChatGPT for the answer is fast and you learn nothing. Socratic agents are slow and you learn something. Use the Socratic pattern for the core concept; use plain chat for the syntax lookup.
• Reflection skill vs writing a journal yourself — both work, but the friction of staring at a blank journal kills most people in week two. An agent that asks "what was the hardest part of today's session?" gets answers a blank page doesn't.
• Fabric create_quiz vs a real practice test — generated quizzes are great for exposing what you don't know (and free, and infinite). Real past papers (for certification exams especially) are the ground truth. Generate quizzes daily; do one official practice test per week.

Common pitfalls (passive reading + over-trusting the AI)

• Letting the AI write the code / essay / proof and calling it learning — this is the #1 failure mode of AI-assisted study. The moment you stop typing answers yourself, learning stops. The Demonstrate Understanding agent (#4) is in the pack specifically to enforce "you say it first, I'll critique."
• Skipping the active recall step — re-reading feels like studying and isn't. Generate a quiz with Fabric create_quiz before re-reading the chapter — if you score 12/15 cold, you didn't need to re-read. If you score 4/15, you know exactly where to focus.
• Hallucinated facts in a domain you can't yet verify — when you're 3 days into a topic, you can't tell if the AI is wrong. Always run DeepTutor or NotebookLM on a real source you trust for factual claims; use the chat tutor for intuition and worked examples.
• No persistent memory = no real personalization — without Mem0 (or equivalent), the tutor treats you like a stranger every session. By week three you'll be bored. Wire memory in day one, even if you have to fake it (export the chat, paste a summary back).
• No reflection = no consolidation — what you don't review in some form within 24 hours, you mostly lose. Claude Reflect at the end of every study block is the cheapest unlock in this pack. Five minutes, every night.
• Trying to install all ten on day one — minimum viable kit is three: Mr. Ranedeer (paste a prompt, 30 seconds), DeepTutor (upload one PDF), and Demonstrate Understanding (one Claude Code agent). That's enough for the first two weeks. Add the rest as you hit the failure modes they fix.