Why I Built CleverYou — And What I Learned

The Problem I Kept Running Into

Two things were broken at once, and I kept running into both. On one side, everyone I know is now learning with AI in an open chat window — ask Claude or GPT to explain something, get a genuinely great answer, close the tab an hour later, and watch most of it evaporate. Nothing forces it to stick. On the other side, the platforms actually built for learning, Khan Academy and its peers, serve a real purpose for a lot of people, but the format is fundamentally old-school: watch a video, do the exercises, watch the next one. Hours of one-directional content, with no real test of how much of it you can actually recall afterward.

That's the problem I kept running into myself. Neither side had it right. Open-ended AI conversation gave me understanding with nothing to hold onto afterward. Structured video platforms gave me structure with nothing adaptive or conversational about it. I wanted a personal tutor that combined the best of both — genuine, adaptive conversation, but built to actually retain what you'd covered instead of letting it evaporate. That gap is what I built CleverYou to close.

What CleverYou Is

CleverYou is built around two tools working toward one goal: making learning with AI actually stick, not just feel good in the moment.

CleverCoaches are voice-first AI tutors, and I'll say it plainly: the adaptive coaching underneath them is groundbreaking. Not a scripted flow with a friendly voice on top — a system that actually changes what it says and how it says it, in real time, based on how you're doing. You talk instead of type, and a coach adapts its explanations as you go, asking questions back and giving live feedback across 8 subjects: Math, Physics, History, Coding, English, Economics, Philosophy, Business, each with its own subject-specialized coach. Kodey handles coding logic and debugging. Atlas walks through historical thinking, timelines, causes and consequences. Elara works on vocabulary, grammar, and conversational fluency. Every session ends with an AI-generated recap and a quiz, so the conversation turns into something you can actually review later instead of evaporating the moment you close the tab.

CleverNotes solves the other half of the problem: the material you already have but haven't turned into anything useful yet. The idea was to give people a real workspace for the sources they already know matter to them, instead of forcing everything through a rigid course structure someone else designed. Drop in a URL, a YouTube video, a PDF, or your own notes, and CleverNotes ingests it and gives you a set of actions to run against it — a Smart Summary, a Study Guide, an Auto Quiz, Flashcards, an Exam Brief, all exportable as a PDF — generated and grounded strictly in what you gave it. No invention, no filler, nothing added that wasn't in the source.

The one output I'm genuinely proud of is what CleverNotes calls the Load-Bearing Concept. Every set of notes comes with the single insight a great teacher would tell you before the lesson even starts, not a summary, not a list of key points, but the one idea that, once you actually have it, makes everything else in the source feel obvious. It's specific to your exact material every time, never a generic template — that's the difference between a tool that summarizes and a tool that teaches.

There's also Clever Debates — a lighter tool built for practicing argument and debate against an AI opponent. It's not the core of the platform, but it's a genuinely nice one to have for anyone who learns by being pushed to defend a position out loud, not just read about it.

The whole platform is built around continuity. Coaches remember exactly where a session left off, notebooks and generated study guides stay saved, and a learning streak tracks how consistently you're actually showing up, because the thing that kills most self-directed learning isn't a bad explanation, it's nothing forcing you to come back tomorrow. There's a free plan to start, no credit card required.

Why I Built It

The real reason goes back further than a gap in my own workflow. This is a generation where people learn in genuinely different ways, and the method that worked for one person in school rarely worked for the person sitting next to them. The idea behind CleverCoaches specifically was simple: everyone deserves a personal tutor who's actually available when they need one, not whenever a human tutor's schedule allows it, and not at the exorbitant hourly rate that puts real one-on-one tutoring out of reach for most people. Voice, real-time, adaptive, and there at 1am the night before an exam if that's when you actually need it — that was the point.

Some of it was personal too. I'm Gabriel, the founder, and Khan Academy is where I did a lot of my own SAT prep. It helped. I'm not going to pretend it didn't. But it didn't help enough — the videos were fine, the practice sets were fine, and none of it adapted to where I was actually stuck or gave me anything back beyond a right or wrong when I missed a question. What I wanted was a person there instead of a library of videos, and there wasn't one, especially not at 11pm two days before a test. That's the specific gap CleverCoaches is built to close.

I didn't start DTD in a boardroom, and CleverYou didn't start as a plan either. DTD began with building websites that gave real businesses a real voice online — people doing the actual work, not the ones with the big budgets. Somewhere in that, I asked myself a harder question: what if I could give anyone access to the kind of learning that actually changes outcomes, not a course, not a video library, but something that listens, adapts, and teaches the way a great human tutor would, at a fraction of the cost, available any time? Access to a good education shouldn't depend on your background, your budget, or your postcode. That's the belief CleverYou got built on.

None of that means the structured, self-paced format Khan Academy built is wrong, just incomplete on its own. There's a real place for it, and it's part of what's coming next for CleverYou: a more traditional, guided lesson format sitting alongside the conversational coaching, for the people who learn better that way.

How It's Built — The Technical Layer

This isn't a no-code wrapper around an API, and I'm keeping the exact stack to myself — but the shape of it is worth describing. Under the hood there's a proper database layer with row-level security so one user's data is never reachable by another, a real authentication and session system instead of a bolted-on login form, a voice pipeline tuned specifically for low-latency, real-time conversation rather than a generic chat widget, and a billing system built to handle real subscriptions and real edge cases, not a payment button glued onto a landing page.

The part that actually took the longest wasn't any of that infrastructure — it was the prompt architecture behind both products. Months went into the coaches. CleverNotes took even longer: the better part of a year refining the layer that decides what actually counts as a Load-Bearing Concept in a given source, instead of just producing a plausible-sounding summary. Every subject-specialized coach runs on a highly structured, heavily iterated set of instructions fed into the underlying language model and voice layer, rewritten and re-tested until a session actually felt like being taught by someone who knows the material, not like talking to a chatbot that happens to know facts. CleverNotes runs on its own equally structured set, tuned until it could reliably tell the difference between a genuinely load-bearing idea and an interesting but secondary detail, source after source, without falling back on a generic template. Getting either of those right, consistently, is a much harder problem than wiring up an API call. Maybe that's ego talking, but I'll stand behind it: what the coaching prompt actually does is groundbreaking, and anyone who's seen the raw thing would say the same. Those two prompt layers are the parts of CleverYou I'm proudest of, and also the parts I talk about the least.

I'm not describing any of this to impress anyone — I'm saying it because it matters for anyone evaluating whether "AI product" means a real system with real auth, billing, data boundaries, and months of prompt work behind the scenes, or a demo with a nice UI in front of a single API call. CleverYou is the former, and the specifics of how I built it are details I'd rather keep close.

What I Learned

A few things became obvious once real users were on the platform, none of which I expected going in.

• +Voice AI UX is harder than it looks. Latency kills a tutoring conversation fast — a half-second delay that's invisible in a text chat feels like dead air when someone's talking to you, and I spent real time tuning around that rather than treating it as solved.
• +The hardest part wasn't the AI. It was making the product feel intentional instead of "AI wrapped around existing content." Anyone can pipe a PDF through a model and get a summary back. Making that summary actually useful for retention took far more product thinking than model prompting.
• +Billing and auth took longer than any AI feature I built. Nobody talks about this part, but wiring authentication, data permissions, and billing together correctly ate more time than the voice tutor did.
• +Users want structure. Open-ended AI is overwhelming — a blank box and "ask me anything" is a worse experience than a guided flow that tells you what to do next. Guided AI, not open-ended AI, is what actually gets used.

Where CleverYou Is Going

More coaches across more subjects, a mobile app so the voice tutor isn't tied to a desktop session, and community features so people studying the same material can actually learn alongside each other instead of in isolation. Nothing here is vague roadmap filler — it's the direct next step from what I've already seen people actually use.

Closing

This is the kind of product I build at DTD — AI that solves a real, specific problem, not AI for its own sake. CleverYou is one example; Zrem is another. If you're building something similar, or thinking about what AI-native product development actually looks like once you get past the demo stage, I'd like to talk about it.