Designing a Learning System Around How My Brain Actually Works

I have ADHD. My attention doesn't deplete gradually. It cliff-drops. One minute I'm locked in, the next I'm reading the same paragraph for the fourth time without absorbing a single word. Most learning tools are built for people who can study for two hours straight. That has never been me, so I built one that doesn't assume it.

25 minutes, hard stop

The session length is not a suggestion. The system ends the session at 25 minutes regardless of where I am. Mid-sentence, mid-concept, mid-question. Done.

This is borrowed from Pomodoro, but the motivation is different. Pomodoro uses timed intervals for productivity. I use them for protection. ADHD brains overshoot. What feels like "I'll just finish this section" becomes 90 minutes of unfocused reading where I retain nothing. The dangerous part is that it feels productive in the moment. It isn't.

The hard stop protects against my own momentum. It forces me to stop while the learning is still good, not after it's degraded into screen-staring. Twenty-five minutes of actual engagement beats two hours of diminishing returns every time.

Retrieval before revelation

The system tests me before showing me the material. "What do you remember about spaced repetition?" before "Here's the chapter on spaced repetition."

This feels counterintuitive. Why test someone on material they haven't reviewed yet? Because retrieval practice, the act of trying to remember before re-reading, produces stronger memory formation than re-reading alone. This is well-documented in cognitive science. The effort of recall strengthens the neural pathway more than passive recognition does.

For ADHD brains, there's a second benefit: urgency. Answering a question is an active task. It demands a response. Reading a textbook chapter is a passive task, and passive tasks are where ADHD attention goes to die. By front-loading retrieval, every session starts with engagement instead of hoping engagement will arrive on its own.

Forgiving scheduling

Miss a day? The system reschedules without comment. No streaks. No "you broke your 14-day chain." No sad emoji. No motivational guilt trip.

This was a deliberate design decision, not a missing feature. Streak-based motivation is catastrophic for ADHD. Here's why: neurotypical learners miss a day and think "I'll get back to it tomorrow." ADHD learners miss a day, see the broken streak, and think "I already failed, why bother restarting?" One missed day destroys the entire motivational framework. The tool designed to keep you engaged becomes the reason you disengage.

Instead, the system uses spaced repetition with the SM-2 algorithm. It adjusts review intervals based on actual recall performance, not attendance. If I remember something well, the interval extends. If I don't, it shortens. The algorithm doesn't care whether I showed up yesterday. It only cares whether I remember the material right now. That's a better measure of learning than any streak counter.

Conversation-driven, not screen-driven

The learning happens through dialogue with an AI tutor. I type or talk. The AI responds with Socratic questions, pushing me to articulate concepts in my own words rather than recognizing them in someone else's.

This isn't a flashcard app with a chatbot bolted on. The conversation is the primary interface. No card decks to browse, no progress screens to check, no settings to fiddle with. Just a conversation.

I chose this because it matches how I actually process information. I don't learn by staring at cards. I learn by explaining things, arguing with ideas, finding the edges of my understanding through dialogue. Flashcard interfaces assume learning is about recognition: "Did you know this? Yes/No." Conversation assumes learning is about construction: "Explain this in your own words. Now, what happens if we change this variable?"

The conversational format also eliminates the paradox of choice that plagues most learning apps. No home screen with twelve options. No "what should I study today?" decision. The system starts talking. I respond. That's it.

The broader point

I designed this system for myself. It is not a product. It is not scalable. It solves one person's problem in a way that would need rethinking to solve anyone else's.

But the design decisions reflect a principle worth keeping: the best tool is the one that works with your actual cognitive style, not the one that works "in general." General-purpose tools optimize for an average learner who doesn't exist. Every real person has specific failure modes and engagement triggers. Designing for those specifics, even at a population of one, beats designing for everyone and therefore nobody.

If you find yourself fighting a learning tool instead of using it, that's not a discipline problem. That's a fit problem. Design around how you actually work, not how the tool assumes you do.