I got an email last week that made my entire year. It was from a mother in Ohio. Her son, Leo, has ADHD and has always struggled in a traditional classroom. He'd been using our AI tutor for a month. She wrote, “For the first time, Leo doesn’t feel stupid. He feels seen. Your tutor knows when he’s getting frustrated before he even does.”
That’s it. That’s the whole reason we started this. People ask me how our AI tutor feels so... human. It wasn't a simple process. There was no magic wand. It was a brutal, exhilarating, and deeply personal journey into the heart of how we learn. This is the story of how we built it.
The Broken Promise of EdTech
I’ve seen hundreds of EdTech pitches. I’ve invested in a few. Most of them are just glorified digital textbooks. They’re static. They’re one-size-fits-all. They throw information at a student and hope some of it sticks. That’s not learning. That’s a content dump.
I remember my own experience with coding bootcamps. You’re in a room with 30 other people, all with different backgrounds and learning speeds. The instructor has to teach to the middle. The fast learners get bored and the slow learners get left behind. I saw it happen over and over again. It’s a fundamentally broken model.
We didn’t just want to build another learning app. We wanted to build a mentor. A guide. Something that could replicate the magic of a one-on-one tutor who truly understands you. Someone who knows when to push, when to pull back, and when to just offer a word of encouragement.
The Ghost in the Machine
The big idea came from an unlikely place: one of my early angel investments, a gaming company. They were using biometric data from controllers to adjust the difficulty of the game in real-time. If your heart rate spiked, the game would ease up. If you were cruising, it would throw a new challenge at you. It was brilliant. It kept players in that perfect state of flow, always challenged but never overwhelmed.
What if we could do that for learning? What if we could read a student’s cognitive and emotional state and adapt the lesson in real-time? That was the spark. That was the moment we knew we were onto something big.
We assembled a small, scrappy team. A couple of data scientists who thought I was crazy, a UX designer who specialized in cognitive psychology, and me. Our first office was a converted garage. The coffee was terrible, but the energy was electric.
Building the Engine of Empathy
Our first challenge was data. How do you measure frustration? Or boredom? Or that "aha!" moment of understanding? We couldn’t use heart rate monitors. We had to rely on what we could see: mouse movements, typing speed, hesitation, deletion rates. We called it “digital body language.”
We spent months just collecting data. We built a simple math game and had hundreds of students play it. We recorded every single interaction. We had them self-report their emotional state every 5 minutes. It was a mountain of data. A messy, beautiful mountain.
Then came the hard part: finding the signal in the noise. Our first algorithm was a disaster. It was basically a random number generator. It would think a student was frustrated when they were just scratching their nose. We almost gave up. I remember one night, at 3 AM, staring at a wall of code, thinking this was the dumbest idea I’d ever had.
The breakthrough came when we started looking at sequences of actions, not just individual data points. A student who deletes a line of code once is probably just fixing a typo. A student who deletes the same line of code five times in a minute is stuck. A student whose mouse is hovering over the “hint” button for 10 seconds is hesitating. It’s the patterns. It’s always the patterns.
We built a recurrent neural network (RNN) that could learn these patterns. We fed it our mountain of data. And slowly, painfully, it started to work. It could predict a student’s self-reported emotional state with 85% accuracy. It was like seeing a ghost in the machine. For the first time, our tutor had a sense of empathy.
More Than Just an Algorithm
But a great tutor is more than just an empathy engine. They’re a master storyteller. They know how to frame a concept in a way that clicks. So we built a content generation system that could create thousands of variations of the same lesson.
If a student is struggling with a concept, our tutor doesn’t just repeat the same explanation. It tries a new one. It uses an analogy. It shows a visual. It breaks the problem down into smaller steps. It keeps trying until it finds something that works. It’s relentless.
We also learned that timing is everything. You can’t just jump in with a hint the second a student gets stuck. That’s annoying. It’s like someone trying to finish your sentences for you. Our tutor learned to wait. To give the student space to struggle. That struggle is where the real learning happens. The tutor only intervenes when it detects that the student is on the verge of giving up.
The Future is Personal
We’re still just scratching the surface of what’s possible. We’re experimenting with voice analysis to detect a student’s tone. We’re building a system that can generate personalized encouragement based on a student’s past successes. The goal is to create a learning experience that is as unique as the student themselves.
This isn’t about replacing teachers. It’s about giving them superpowers. It’s about automating the personalized instruction that every teacher wishes they had the time to provide. It’s about making sure that no student, like Leo, ever feels stupid or unseen again.
Building this has been the hardest thing I’ve ever done. It’s been a journey of a thousand tiny failures and a few, glorious breakthroughs. And we’re just getting started.
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.