skip to content
Navid Mesbah
فا
← Back

Teaching and the Design of Interactive Explainers

/ 10 min read

Where Understanding Begins

For many years I have worked with learners, as a teaching assistant at university and as a tech and computer science coach. It started with mathematics, and over time it grew into computer science, systems and architecture, software engineering, technology, business, and more. University sections, workshops, small groups, one-on-one sessions, late-night help before exams, and long conversations with founders and engineers trying to understand a hard system. Students who loved the subject and students who were sure they hated it.

If I had to name the one thing I learned, it’s that understanding happens in a moment you can see.

A student’s face changes. Their eyes stop moving between the board and their notebook and settle on something in the middle distance. They stop copying and start predicting. They say “wait, so if…” and then finish the sentence correctly before you do. It’s one of the most beautiful things I’ve ever witnessed, and I saw it hundreds of times, and it never got old.

Almost always, that moment comes from seeing something move.

A graph that bends as you change a parameter, and suddenly the student sees what the parameter does instead of memorizing it. A triangle whose angles still add up to 180 degrees no matter how you drag its corners, and suddenly the rule feels inevitable instead of arbitrary. A pendulum that slows near the top of its swing and speeds up at the bottom, and suddenly energy conservation is not an equation but a thing you can feel.

The formula comes after. The intuition comes from motion.

What I Learned About How People Learn

Years of coaching across these fields taught me some things that no education course ever did. A few of them shaped everything I’m building now.

Most confusion is about order, not difficulty. When students struggled, it was rarely because the idea was too hard. Usually, it was because they had met the ideas in the wrong order. They were asked to use a concept before they had a picture of it, or shown a formula before they had a reason to want it. The same material, rearranged, often became easy.

People learn from predictions. The most powerful teaching move I know is to stop before a reveal and ask: “What do you think will happen?” A student who predicts, even wrongly, is engaged in a way that a student who watches is not. A wrong prediction followed by the right answer is often more memorable than a correct explanation.

Mistakes are information. When a student makes an error, it tells you exactly how they’re thinking. A good teacher reads mistakes like a doctor reads symptoms. The worst thing you can do is simply mark it wrong; the best is to ask why it seemed right.

Shame kills learning. Many students who “hate math” were simply made to feel stupid at some point, and they never recovered. Once a student believes they’re bad at math, every confusion confirms it, and they stop trying. Rebuilding confidence was often my first job, before any actual mathematics.

Every student needs a different entry point. Some students needed pictures. Some needed stories. Some needed to touch and move things. Some needed the formal definition first and found intuitions confusing. The same idea could be explained in five ways, and each student would understand one of them best.

Interaction beats explanation. I could explain something beautifully and see blank faces. Then I’d hand the student a ruler, a piece of string, or a graphing tool and let them try, and the blank faces would light up. Explaining tells. Interacting shows. Showing is stronger.

The Problem With AI Explanations

Today anyone can ask a chatbot to explain a derivative. The explanation is usually fine. Accurate, clear, well-organized.

It’s also usually a wall of text with a formula in the middle, which is exactly how most students failed to learn it the first time.

This is the strange situation of AI in education right now. We have systems that can explain almost anything in words, patiently and at any hour, in many languages. That’s genuinely valuable, and I don’t want to dismiss it. A student stuck on homework at midnight can get help that was never available before.

But words are not how most people learn mathematics. Mathematics is fundamentally about structure, relationship, and change, and those are things you see before you can say. A text explanation of a derivative is a description of a picture the student has never seen.

And when AI does try to make visuals, accuracy suffers. A wrong graph is worse than no graph, because students trust pictures more than words. I’ve seen generated diagrams with curves that don’t match their equations, vectors pointing the wrong way, triangles labeled with angles that couldn’t possibly be correct, and physical simulations that violate the physics they’re supposed to illustrate. To a student who doesn’t yet know the subject, these errors are invisible. They absorb them as truth.

For education, “almost right” is not acceptable. A teacher who draws a wrong graph on the board loses the class’s trust. An AI that does it loses something worse: the student’s correct understanding.

What I’m Building

That’s why I’m building an engine that turns a prompt into a cinematic, accurate, interactive explainer for math and science.

A student or teacher types something like “explain why the derivative of sine is cosine” or “show me how a projectile’s path depends on launch angle.” The engine produces not a paragraph, but a short, directed, visual experience: an explanation that unfolds with pacing, animated mathematics that moves, and interactive elements the student can play with.

It’s Persian-first, because Persian-speaking students are who I coached and who I most want to serve. But the principles are universal, and I believe they would help students in any language.

Three words define what the engine is trying to do.

Cinematic means the explanation has pacing and direction, like a good teacher at the board. A good explanation is not a dump of information; it’s a performance with a shape. It builds up, introducing one idea at a time. It pauses, giving the student a moment to predict. It reveals, showing the answer at the moment it will have the most impact. It returns, connecting the new idea to what the student already knew.

Teachers do this intuitively. Good educational videos do it deliberately. The engine plans an explanation as a sequence of beats, the way a film director plans scenes, before anything is drawn.

Accurate means the visuals come from real computation, not from imagination. This is the most important principle and the one I’m least willing to compromise on.

When the engine shows the graph of a function, the curve on screen is computed from the function. When it simulates a pendulum, the simulation obeys the actual equations of motion. When it labels an angle, the label is calculated from the geometry. Language models plan the explanation and write the narration, but the mathematics is done by mathematics.

We also check generated content against the math, not against what looks plausible. If the narration says “the curve crosses the axis at x equals 2,” the engine verifies that it does. If an explanation claims a relationship, the engine tests it numerically. Errors that slip through are treated as serious bugs, not acceptable imperfections.

Interactive means the student can grab things. Change a parameter and watch the graph respond. Drag a point and see how the triangle’s angles adjust. Pause the simulation, rewind, try a different initial condition. The goal is for students to ask “what if?” and get an answer immediately.

Interaction is where curiosity becomes learning. A student who wonders “what happens if the angle is 90 degrees?” and can try it right away learns more in that ten seconds than from a page of explanation. And because they asked the question themselves, they remember the answer.

What Coaching Taught Me About Building

Many of the design decisions in the engine come directly from lessons in the classroom.

Plan the order first. Because most confusion is about order, the engine spends real effort planning the sequence of an explainer before generating any visuals or text. What does the student need to see first? What intuition must be in place before the formula? Where’s the natural moment to ask for a prediction?

Show the wrong path too. Students learn from mistakes, not from perfect demonstrations. The best explainers anticipate common misconceptions and address them directly: “You might expect the heavier ball to fall faster. Let’s see.” The engine includes common misconceptions in its planning and, where appropriate, lets the student make the tempting wrong prediction before showing what actually happens.

Offer multiple entry points. The same idea can be explained geometrically, algebraically, physically, or through a story. The engine can produce different versions for different learners, and students can switch between them. Some will prefer the picture; some will prefer the equation. Both are valid paths to the same understanding.

Never make anyone feel stupid. The tone of an explanation matters as much as its content. The engine is designed to be warm, patient, and encouraging. It never says “obviously” or “simply” or “as everyone knows.” Those words have hurt more students than any difficult theorem.

Respect the learner’s intelligence. At the same time, students can tell when they’re being talked down to. A good explanation trusts the student to think. It asks real questions and leaves room for real discovery.

The Role of the Teacher

People sometimes ask whether tools like this will replace teachers. I don’t think so, and I don’t want them to.

A good teacher does much more than explain. They notice when a student is struggling before the student says anything. They know which student needs encouragement and which needs a challenge. They build a classroom culture where questions are welcome and mistakes are safe. They inspire. They care. No software does these things well, and I’m skeptical any software will soon.

What I hope is that tools like this can give teachers better material. A teacher who wants to show a class how a sine wave relates to a rotating circle shouldn’t need to spend hours building an animation. They should be able to describe what they want and get an accurate, interactive explainer in minutes, then spend their time on what only they can do: teaching the actual students in front of them.

And for the many students who don’t have a great teacher, which is sadly a very large number, I hope these tools can carry at least part of what happens in a good classroom: the pacing, the pictures, the chance to explore, the patience.

Why This Matters to Me

When I think about the students I worked with, I remember the ones who had that moment of understanding. But I also remember the ones who didn’t. Students who were capable but never got the right explanation at the right time. Students who decided math wasn’t for them and carried that belief into the rest of their lives, avoiding careers and opportunities because of a subject they could have loved.

I think about how different it might have been for them if they’d had a way to see the ideas move, to try things, to make predictions and test them, in their own language, at their own pace.

More than a thousand students over the years is a lot of faces. But it’s a tiny number compared to the millions of students who could learn this way. If a tool can carry even part of what happens in a good classroom to a student who never had a good teacher, it’s worth building.

And if, somewhere, a student who was sure they hated math drags a slider, watches a curve bend, and says “wait, so if…” then everything I’ve worked on will have been worth it.