I’ve always been jealous of people who can casually explain satellite telemetry or quantum computing at a dinner party. Not in a show-off way — in a “this person genuinely understands something deeply” way.
So I built a system to turn myself into one of those people. One subject at a time.
The Idea
MakeMeAnExpert (MMAE) started as a simple cron job. Every morning at 6 AM, I get an email with a lesson on a subject I want to master. Right now it’s “AI/ML in Satellites and Space Systems” — because why start easy?
The system generates a 30-day syllabus that ramps from fundamentals to advanced topics. Day 1 might cover orbital mechanics basics. Day 15 digs into onboard ML inference constraints. Day 30 ties everything together with current research frontiers.
But here’s the thing — just reading something every morning doesn’t make you an expert. Passive consumption is how you end up with 47 saved articles you’ll never revisit. I needed the system to force active engagement.
The Science of Actually Remembering Things
Before building the learning features, I went down a research rabbit hole. Turns out, cognitive science has some strong opinions about what works and what doesn’t.
Dunlosky et al. (2013) reviewed hundreds of learning studies and ranked techniques by effectiveness. The winners?
| Technique | Effectiveness | Why |
|---|---|---|
| Practice testing | HIGH | Forces retrieval, strengthens memory pathways |
| Distributed practice | HIGH | Spacing beats cramming every time |
| Self-explanation | MODERATE | Making yourself articulate “why” deepens understanding |
| Highlighting/re-reading | LOW | Feels productive, isn’t |
The Feynman Technique — explaining something in simple terms as if teaching a child — aligns with several of these. It forces retrieval (you can’t explain what you don’t know), self-explanation (you have to articulate the “why”), and exposes gaps (the moment you stumble, you’ve found what to study next).
So I baked all of this into the system.
What a Daily Lesson Looks Like
Every morning’s email isn’t just a wall of text. It’s structured to maximize retention:
1. The Lesson — 10-15 minutes of content on the day’s topic. Real explanations, not Wikipedia summaries. Includes images from NASA and Wikipedia with proper attribution.
2. Prediction Prompt — Before the lesson even starts, I’m asked to predict what I think the answer to a key question will be. Research shows that making predictions (even wrong ones) dramatically improves learning because your brain pays more attention to the correction.
3. Callback Question — A question that connects today’s lesson to a previous day. “How does today’s topic on onboard processing relate to the orbital mechanics you learned on Day 3?” This is spaced repetition in disguise.
4. Feynman Prompt — “Explain [today’s concept] in simple terms, as if teaching someone with no technical background.” I write my explanation in a markdown file in the GitHub repo, and an AI grading system reviews it.
5. Weekly Quizzes — Days 7, 14, 21, and 30 include cumulative quizzes. The testing effect (Roediger & Karpicke, 2006) shows that retrieval practice improves long-term retention by ~21% compared to re-studying.
The Grading System
This is where it gets fun. Each lesson auto-generates answer template files in the repo:
answers/
day-01.md ← prompts auto-populated when email sends
day-02.md
...
quiz-week-1.mdI fill in my answers directly in GitHub (or clone and edit locally), then run:
./grading/grade.sh day-01The grading script feeds my answer plus the original lesson context to Claude, which evaluates understanding on multiple dimensions — accuracy, depth, ability to connect concepts — and emails me feedback with a score and suggestions for what to review.
It’s like having a patient professor who’s available at 11 PM and never judges you for confusing azimuth with altitude.
Building Toward an App
The cron-job-and-GitHub setup works surprisingly well for a single user (me). But I got excited enough about the concept that I started building a proper web app around it.
The architecture:
- FastAPI backend with PostgreSQL — user accounts, subjects, lessons, enrollment tracking, answer submission, AI grading
- Next.js frontend — dark “scholar” theme, subject catalog, lesson viewer, progress dashboard with streak tracking
- AI grading pipeline — answers evaluated by Claude for understanding depth, not just keyword matching
The idea is that anyone could sign up, pick a subject, and get the same structured 30-day deep-dive experience. The system pre-generates lesson content so there’s no waiting, and the grading adapts to your level.
Lessons from Building a Learning System
A few things surprised me while building this:
Images matter more than I expected. Adding NASA imagery and Wikipedia diagrams to lessons made them feel substantially more real. A lesson on synthetic aperture radar hits different when you can see an actual SAR image of a coastline.
The prediction prompt is secretly the best feature. It feels trivial — “what do you think X does before reading the answer?” — but it consistently makes me pay closer attention to the actual lesson. My brain treats the lesson as a correction rather than new information, which apparently sticks better.
Weekly quizzes are genuinely stressful in a productive way. By Day 14, the quiz covers two weeks of material and I consistently discover gaps I thought I’d filled. That’s the whole point.
Archiving lessons to a separate repo was a small decision with big value. Having a searchable markdown archive of everything I’ve learned means I can reference Day 4’s orbital mechanics when I’m struggling with Day 18’s GPS signal processing.
What’s Next
The first 30-day cycle (AI/ML in Satellites) wraps at the end of March. After that I’m eyeing a few more subjects:
- Distributed systems design
- Computational neuroscience
- History of cryptography
The web app is in active development — backend scaffolding and frontend are done, core API endpoints are being built now. The goal is to have it deployed by mid-March so other people can try it.
If you want to follow along with my satellite AI journey, the daily lessons are being archived at ai-ml-satellites as they generate.
The honest truth is that I built this because I was tired of feeling like I only kind of understood things. Surface-level knowledge is easy to accumulate and useless to apply. MMAE is my attempt to go deeper — one structured, quiz-enforced, Feynman-prompted day at a time.
And hey, I can already explain what a sun-synchronous orbit is at dinner parties. Progress.