Ritsu is the AI tutor that trains you to deeply understand, practice, retain, and apply whatever you're learning. It's for the moment an explanation feels clear and a practice question proves it wasn't: it turns your own chapter, slide deck, lecture video or paper into an ordered plan, asks you to explain each idea back in your own words, holds back a problem's solution until you try, and brings back what starts to fade. It launched October 19, 2026.
Most students already have something that explains things well. An engineering student described what studying with a chatbot did to them: "gemini explanation is often detailed and I feel like I understand well but since using this method it feels like my grades got even worse" (thread). The explanations were good, and that may be the problem: producing the answer yourself is the part of studying that does the learning, and a good explanation leaves you nothing to produce.
What is Ritsu, and what does "makes you do the learning" mean?
Ritsu is a study system built around your file, and "makes you do the learning" means the tutor gets the explanation out of you instead of handing one to you. Reading a clear explanation, even a perfect one, is something that happens to you. Writing your own explanation, solving a problem or catching a mistake is something you do, and that is the work learning is made of.
Michelene Chi and Ruth Wylie (2014) put this difference on a scale. Sort study activities by what you visibly do, they proposed, and learning tends to rise with each step: passive (reading, listening), active (highlighting, copying, rewinding), constructive (your own explanation, a concept map, a solved problem) and interactive (building that explanation with someone who pushes back). They tested this ICAP hypothesis against lab and classroom studies of note-taking, concept mapping and self-explaining, so read it as a strong default, not a law of nature.
Here is where a good chatbot falls short. Reading its explanation, however good, is passive, and the better the explanation, the less of it you have to make yourself. Understanding something means you can rebuild it without the page: say it in your words, use it on a problem you haven't seen, and notice when an answer is wrong. Recognizing it when you see it again is a different and much weaker skill, and it is the one rereading trains.
If that sounds like your last study session, try the other order on this week's chapter or lecture: Ritsu builds the plan from your file and asks for your explanation before you move on. Start Learning Free
Why does active learning feel wrong when it works better?
Because effort feels like failure while it is happening. In a 2019 study at Harvard, Louis Deslauriers and colleagues taught the same introductory physics material, with the same handouts, two ways: as a polished lecture in some class sessions, and in others as problem-solving sessions, where students worked through problems in small groups and the instructor stepped in with feedback. Education researchers call that active learning; on Chi and Wylie's scale it is constructive and interactive. On the test that followed, students did better after the problem-solving sessions. Asked how much they had learned, they rated those sessions lower.
The authors traced the gap, in part, to effort. Working through a problem feels clumsy; a fluent lecture feels clear. Students read the clumsiness as a sign they weren't learning. The study ran in two introductory physics courses at one university, so it doesn't settle every subject, but the pattern is one you can check on yourself tonight.
Call it the fluency trap: the material feels clear, so you assume you have it. That feeling measures how easy the page is to follow right now, not whether you can produce it next week.
Try it on the last thing you studied. Close your notes and write the main idea of the last section in three sentences, then open them and compare. Whatever you could only recognize, and not write, is where tomorrow's session starts.
AI tutor vs AI chatbot: what's the difference?
A chatbot answers what you ask, in the order you ask it. A tutor decides what to ask you, in an order that builds, checks your answer against the material, and comes back to what you got wrong.
| A chatbot, by default | A tutor | |
|---|---|---|
| Who picks the next question | you | the tutor, in the order the material builds |
| When you're wrong | it explains the right answer | it shows what's missing and asks you to try again |
| Where the answer comes from | its training, unless you paste a source | your material, with the page it came from |
| Next week | nothing happens | what you missed comes back for review |
You can push a chatbot toward the right-hand column with rules set at the start of a chat. They hold until the night you're tired and type "just explain it", and a chatbot is built to say yes (why answer-first AI hurts learning).
What does a study session need to do?
Four jobs, in order, with a notebook or any tool; each ends with something you can check.
- Plan (10 minutes). Break the material into small units in the order they depend on each other. You're done when you can name the first unit and say why it comes first. In Ritsu: your file becomes an ordered path of small units, every item traced to the page it came from.
- Learn by explaining (15 minutes a unit). Before you read an explanation of a unit, try to write one. Then read, then explain it again in your own words, without scrolling up. You're done when your second explanation names something your first one missed. The Feynman technique is this step done on purpose. In Ritsu: /explain gives you an explanation when you ask for one, and the tutor then asks you to explain the idea back in your own words.
- Practice with feedback (15 minutes). Answer questions and solve problems with the solution closed, then find out exactly what was wrong, not only that it was wrong. You're done when a question from a new angle on the same idea doesn't trip you. That's active recall. In Ritsu: /solve in DIY mode holds the solution back until you write an attempt, and a wrong answer gets the misconception behind it named, with a guided retry.
- Master, then review (5 minutes now). Count a unit as done only when you can retrieve, explain or apply it without your notes, and bring it back after a day, a few days and a week, before it fades. Spaced repetition schedules those returns for you. In Ritsu: every concept gets a score you earn by answering, the Knowledge Map shows what you've mastered and where the gaps are, and review is scheduled from your own file with no cards to make.
What does this look like on one lecture?
Take a lecture on how neural networks learn. The plan, in dependency order: neurons, weights and bias; activation functions; loss and gradient descent; backpropagation; overfitting and regularization. Backpropagation sits fourth because it only makes sense once you know what a loss is and what gradient descent does with a gradient.
Learn by explaining. A first attempt at backpropagation, written before reading, often sounds like this: "The network runs backwards to work out the answer." It sounds right and it's wrong in a useful way. The network still makes its prediction going forward. What travels backwards is the error: starting at the output, the chain rule works out, layer by layer, how much the loss would change if each weight changed. That number is the weight's gradient. The rewritten explanation: "Backpropagation works out, from the output back to the first layer, how much the error would change if each weight changed."
Practice with feedback. Before you read on, answer this one yourself, in one sentence and without options: which step actually changes the weights?
The tempting answer is backpropagation, and it's half right. Backpropagation computes the gradients; gradient descent uses them to change the weights, each one moved a step against its gradient. That step's size is the learning rate, which makes a good recall card: "What does the learning rate set?" Answer: how big each weight update is.
Master, then review. Tomorrow, close the slides and answer three questions from memory: what travels backwards in backpropagation, which step changes the weights, what the learning rate controls. Whatever you miss goes on the list for three days later, then a week.
The explanation came second. What stuck came from the two answers that were wrong.
Who is Ritsu for, and when is it the wrong tool?
Ritsu is strongest where the material is dense and cumulative, the way STEM, pre-med, law, econ and finance courses are: each chapter rests on the last, and there is a lot to read. It is weakest where there is little to read and the learning happens live. On the Free plan a source can be up to 40 pages or 30 minutes of video, so a long textbook goes in chapter by chapter (Split PDF slices one in your browser). And if you can't attempt a unit at all yet, read it once first; the first question needs something to push against.
What does the same session look like in Ritsu?
Every job above works with a notebook and some discipline. The one that breaks first is the second: writing your own explanation right after you've read a good one, every time, when moving on is one click away. A prompt can make a chatbot ask once; keeping it asking at every unit, grading what you write and bringing the questions back next week is on you. Ritsu asks and grades without a reminder, on your own file, and one click on Add to Review brings the questions back on a schedule. Here are the four jobs on the same lecture, from the launch film above.
The moment that matters. In Ritsu the explanation doesn't end the step. Ritsu turns the question around, "Explain backpropagation in your own words.", and grades what you write. In the film, the feedback names the exact idea the answer needed: "You traced the error backward through each layer and named the chain rule." Then the next question checks it from another angle. You find out whether you can produce the idea while the lecture is still open, not on the exam.
The same step on the live app, on a physics chapter: a half-right answer marked Partial, the feedback naming the missing half, a next step to try, and the explanation and model answer last.
1. Drop in the lecture. Drag the file in and press Start Learning. Ritsu reads it and writes the plan unit by unit, each unit traced to its page, then Start opens the first one.
2. Learn, then explain it back. Type /explain with the concept, read the explanation and its figure, then answer the request to explain it in your own words, the moment in the first clip.
3. Practice until it holds. A quiz question checks the same idea from another angle, and a flashcard made from the lecture asks you to recall it and rate how well you did (Forgot, Hard, Good, Easy). Add to Review puts it on a schedule: one day, three, seven, 16 and 35.
4. See what you've mastered. The Knowledge Map lights up concept by concept, mastered ones apart from the gaps, and the plan completes row by row.
A few more things you'll notice on the first day. Every question and explanation shows the page of your file it came from, so you can check the tutor against your own lecture. It runs in the browser on a laptop, tablet or phone, with nothing to install. And Ritsu states that it does not use your data to train AI models. The same flow on a slide deck is on the lecture deck use case.
Bring this week's chapter or lecture. Upload it, let the plan write itself, open the first unit, and when Ritsu asks you to explain it in your own words, write your answer before you scroll back up. Start Learning Free Free forever · No credit card · No time limit
FAQ
Q: What is the ICAP framework?
ICAP is Michelene Chi and Ruth Wylie's (2014) way of sorting study activities by what the learner visibly does: Interactive, Constructive, Active, Passive. It predicts that learning rises from passive to interactive, so writing your own explanation usually beats rereading it.
Q: Why does studying feel productive when I'm not learning?
Because rereading and following explanations feel fluent, and fluency feels like knowing. It only tells you the page is easy to follow right now. Close the page and write the idea out: what you can produce is what you know.
Q: Is explaining something in my own words better than rereading it?
Usually, yes. Explaining makes you build the idea and exposes the step you can't fill in; rereading mostly makes the page familiar. Check your version against the source afterwards: a confident explanation can still be wrong, and fixing it is where much of the learning happens.
Q: How do I know if I've actually learned something?
Run a Mastery Check: a short test you run on yourself before moving on. It doesn't ask "Did I finish?" but "Can I retrieve, explain or apply it without my notes?" If you can do all three with the page closed, move on. If not, that gap is your next session.
Q: Should an AI tutor ever just give me the answer?
Yes, when the answer is a fact you need to look up, a definition, or a step you're still stuck on after a hint. It shouldn't when the answer is the skill you're practicing. Ask for a hint before the answer, and for the answer only after you've tried.
Q: Is there a free AI tutor?
Yes. Several AI tutors have free tiers. Ritsu's Free plan costs $0 with 600 credits a month, needs no credit card and has no time limit; it includes every activity and learning mode, with 40 pages or 30 minutes of video per source. Paid plans add credits and larger sources (pricing).
The takeaway
A good tutor doesn't let an explanation end the step: it asks for yours, because the answer you produce is the one you keep.
Upload this week's chapter or lecture to Ritsu and write your own explanation of the first unit before you open the second. Start Learning Free
Sources
- Chi, M. T. H., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. https://doi.org/10.1080/00461520.2014.965823
- Deslauriers, L., McCarty, L. S., Miller, K., Callaghan, K., & Kestin, G. (2019). Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom. Proceedings of the National Academy of Sciences, 116(39), 19251–19257. https://doi.org/10.1073/pnas.1821936116
- Ritsu pricing page, https://ritsu.ai/pricing (read 2026-10-18)
By Ritsu Team · Last updated October 18, 2026
