I Got Kicked Out of Columbia University for Building an AI That Cheats Interviews - Here’s Exactly What Happened, Why I Did It, and What Comes Next

Quick context before we get into it

Hello, my name is Roy. I founded this company called Cluely and I just got kicked out of Columbia University for building this tool called Interview Coder. It’s like a cheating tool for software engineering, technical interviews. I use the technology to build a much bigger company, Cluely - the desktop app that lets you cheat on everything.

Right now we just launched about few months ago, closing in on $5 million in annually recurring revenue. And we also just closed a $5.3 million seed round led by Abstract Ventures and Suse Ventures.

That’s the headline. But if you only see the headline, you’re going to miss the actual story - the grind, the doubt, the tweets, the hate, the pressure, the product decisions, and why I’m convinced this is a turning point for how humans think and work. So here’s the complete breakdown, exactly as I lived it, no detail left out.

Who I was before all this

I was a pretty - I was pretty wild kid. I got in a lot of trouble. I was pretty smart. I think I was pretty good at math. I did a bit of math competition when I was younger.

When I say “wild,” I don’t mean the cartoon version of rebellious. I mean I was restless, curious, always pushing against rules because I wanted to test where the boundaries were. That restlessness showed up early - in classrooms, in how I argued with teachers, in the kinds of hobbies I picked up and dropped. I wasn’t trying to be a problem child, I just judged things by whether they were fun, challenging, or a game I could win. If it felt like a game, I locked in. If it felt like a chore, I ran.

Math felt like a game. Competitions were a rush. You sit down in a cold room with a pencil, some scratch paper, and questions that look like riddles disguised as math. The timer starts and your brain either lights up or goes blank. Mine lit up. I liked the feeling of seeing a clean trick, spotting a pattern, beating the problem before the time ran out. I wasn’t a prodigy or anything, but I learned to think with structure and speed. I learned how to stare at a question and ask myself: what’s this really testing?

I was on like the debate team. Debate scratched a different itch - it’s chess with people. You stand up, your heart is beating, and you have to think faster than your mouth while also sounding like you’ve been ready for this question your whole life. You learn how to frame points, how to bend a narrative, how to read a room and know when to be sharp and when to be playful. Those tournaments taught me a lot about pressure. They also taught me to stop fearing other people’s judgment. When a judge writes, “confident, aggressive, compelling,” you remember it. When they write, “rambles, unfocused,” you remember that too and you fix it.

I played some cello. That was discipline more than anything else. Practicing scales when all your friends are out, trying to keep a smooth tone when your fingers are numb - it’s not glamorous. But it shaped how I approach hard work. You repeat the motion until your hands remember it better than your brain. There’s a lot of that in coding too - you grind patterns, you iterate, you fix tiny errors until the whole thing sounds smooth.

I love girls. You know, every year I had a new girlfriend. That’s not a flex - it’s just the truth of how social I was, how much I liked connection, how I was always chasing something new. It also meant I was always learning people - what they say, what they don’t say, who needs directness versus gentleness. I didn’t know it then, but that skill would be useful later online when the comments come flying and you need to know which ones matter and which ones are just noise.

Yeah, my mom made me do a lot of studying, but I hated studying. I was always trying to go out and like, like, like, like have fun, play with my friends. She would push textbooks toward me and I would angle toward the door. But she didn’t let go. She kept reminding me that the habits you build when you’re young turn into the person you become. So I studied. Not because I loved it, but because she said, do it, and I listened. And honestly, I’m grateful. That basic discipline got me into good programs, good circles, and gave me a scoreboard to chase.

I’ve always been very competitive. When I was a kid, I had a - I have an older brother who’s two years older than me. I always wanted to be like, smarter than him. I always wanted to do better and great at school than him. Sibling rivalry teaches you speed, it teaches you grit, and it teaches you how to convert jealousy into fuel. I wanted to beat him in grades, in clubs, in anything that had a list or ranking at the end.

I’m always trying to win. I’m trying to be Cheongyu Il Tung at everything. You know, like top of the class in the world of software engineering. That phrase - Cheongyu Il Tung - is how I think about it: number one, top of the heap, first among peers. It’s not about being perfect. It’s about deciding that if there’s a ranking, then I’m going to climb it. If there’s a leaderboard, I’m going to see my name on it. That’s my default setting. It’s not always healthy, but it makes me dangerous when I commit.

Big tech interviews are broken

If you want to get a job at a big tech company, you have to answer these sort of riddle-esque questions that are called leetcode questions. And pretty much every developer you know at a big company has gone through the gauntlet of memorizing 600, 300 riddles and just sort of like memorizing solutions and regurgitating them in interviews.

I, I’m very, very competitive. So the second I knew that there was like a ranking on leetcode, like a global ranking, I knew I had to be like one of the best. So I, I spent hundreds of hours studying, grinding the riddles. Even though I like, I don’t care about leetcode, I didn’t enjoy it. I didn’t really have a good time, but I was just competitive. So I thought like, if there’s a ranking, then I gotta be on the top of the ranking. But I mean, it just ended up with me wasting a bunch of hours.

Leetcode just has nothing to do with what you do on the job. It’s like the modern day equivalent of asking how many balloons fit in the Empire State Building. It’s supposed to test your critical thinking. But the questions are online to the extent that rather than practicing critical thinking, you just practice memorizing the riddles. You’re going to sit through and memorize all the 1000 questions because it means you get a 200k a year job.

This is not good for anybody. You don’t. You don’t learn anything from practicing these riddles, and you just end up wasting time when you should be programming. I thought this was pretty stupid. And this has been going on for around 20 years right now.

The spark: build the tool nobody wanted to admit they needed

The technology was there to sort of develop this tool that would invisibly let you use AI to cheat on these interviews. So I built the tool. I publicly recorded myself using it on the Amazon interview. I got the job, and I posted this everywhere, saying, look how easy it is to hack these interviews. Eventually, this got me in some trouble. But the impetus of everything was when I decided that it’s just a stupid industry practice and I wanted to change it.

Here’s what that looked like in real life. I sat down for the interview with a clean desktop. When the interviewer started talking, I let Interview Coder listen in. The translucent overlay appeared like a ghost layer on my screen - not blocking anything, just floating, whispering context, code patterns, hints, strategies. It wasn’t writing full solutions for me - it was reading the room and helping me decide what to do next, in real time. When the problem landed, the overlay distilled the ask into something actionable: data structure, complexity target, potential pitfalls, edge cases to mention out loud, a skeleton to type. It felt like playing with wallhacks turned on. You still need to aim, you still need to move, but suddenly you can see the whole map.

When I got the yes, I didn’t keep it quiet. I posted it. Everywhere. Not as a humble brag, but as a giant flashing sign: this is gameable. If I can do it, anyone can do it. And if anyone can do it, then the game is broken, and the rules need to change.

The month where nothing happened and everything almost died

A lot of people think it was explosive from the start, but it took about a month before it really started to take off. And for a month, I posted everything. Amazon saw was getting mad at me and Columbia University saw and was getting mad at me, and everyone was just. And it didn’t really go that viral for about a month.

And during that time, I was really stressed. I just gave up my entire career and my entire education for the hope of a company. But it didn’t even go viral. Like, I did all this for like 15,000 views. And I was really worried. Everybody in my life, including even my co founders, were telling me, like, hey, we should probably stop. We should probably shut this down.

But I don’t know. There was just a voice in my head that said, like, this is something. This has potential. Like, I have to keep going. And I did keep going. And then at one point, it did go viral, like super viral. And everybody on tech saw it.

At that point, I was safe. Virality protected me from further punishment from Colombia. It made the path to entrepreneurship a lot easier and clearer. I’ve sort of like, committed my life to building companies. Once I made that decision, it was very easy for me to decide to leave Colombia. And once I did leave Colombia, like, I knew I’m going to do the only thing that I can do now, which is build companies. I think it also helps position me. Like, I’m a very unique person now who got kicked out of an Ivy League. And as a result, there’s a lot of interest from Silicon Valley about what I’m going to do next.

Speed kills. Slowness kills faster

To me at the time, it felt like things were moving really slow. As soon as Interview Coder went viral, I knew that I had to capitalize on the moment because the attention wasn’t going to be there for long. The product was a product designed to die, the product to cheat on technical interviews. So the second companies change technical interviews, the product dies, meaning I have a few short moments before my - my spot in the limelight dies.

And I thought, I have to raise a round. I have to start a bigger company that’s more sustainable and defensible long term. So for me, it felt like things were moving so slow, I had to push back my fundraise. Like two, I was going to fundraise two weeks earlier than I did. To me at the time, I felt like I was in such a time crunch and I had to wrap things up and do the next thing. And it felt like things were moving super slow, but I guess I was under the pressure of the situation.

Yes, parts of it were planned - the rest was reps, reps, reps

Yeah, I mean, we planned for me to use Interview Coder to cheat on a bunch of big tech interviews and get the jobs, and we thought that was going to be a viral moment. So in that sense, we planned it.

But the last three months, I’ve made probably maybe like a thousand tweets. And since then, I’ve figured out how to make tweets that will go viral, how to make tweets that will be more controversial and get more engagement. I think for X especially. Well, I’ve only really cracked X, I think, because I think people on Twitter are a very unique type of people. They love controversy, they love drama, they love attention, and they love to either dog on people or watch people get dogged on.

I think every single time you tweet something, if you don’t think half the people in the world would feel very negative about this, then it’s probably not going to be viral as a tweet. All of your tweets that will, you’re planning on making go viral, they need to have a very strong controversial twist that makes people pause and be like, what the fuck? And this is not the case for Instagram, TikTok or LinkedIn or whatever, but it is the case for Twitter. And yeah, I think, I think Twitter next. The more controversial you’re willing to make a tweet, the better and the more viral you will go.

Hate, love, and why none of it really matters

Interview Coder on Twitter was received so positively. It was just like this scrappy young kid who was trying to fight back against big industry, big tech. And everyone on Twitter was very supportive. I think as I got bigger and my account grew, people grew less supportive, which is like, to be expected. I’m generally very good at receiving hate and criticism. I’m a very polarizing personality and I do a lot of crazy stuff. Throughout my life, I’ve always had people, like, giving me hate. None of the negative comments really stood out, but I was very surprised to see how positively Interview Coder was received on Twitter when I first launched it.

I think people are often so worried that they’re going to say something bad online and it’s just going to get back to them and their reputation is over and I don’t know, like, it’s. It’s going to bury them. But I think in reality, all press is good press. I say a ton of super controversial stuff, and in every video I’m in, there’s like a bunch of comments saying, oh, this guy’s evil. This guy’s like, like this, this. And then there’s always some hate. But it’s like, like this stuff really doesn’t matter.

Yeah, I mean, like, I. I’ve learned that it really doesn’t matter if, if. If everybody in the world just doesn’t like you. Well, actually, it’s. It’s done pretty much the opposite. I’ve realized that even if I say extremely crazy shit online, it will just make people more interested in me and the company, and it’ll just drive more downloads and conversions and get more eyeballs onto Cluely. If anything, I’ve learned that I need to become crazier online so that people will keep funneling attention towards the core product.

I think there’s. It’s very, very rare that you will say something online that translates to something negative happening in the real world. Like, online is not real life. I’m a pretty chill person in real life, but online, I’m crazy because it gets. It gets me engagement and attention.

Split the personas: the online me vs the real me

When you are so polarizing and controversial online, you need to very clearly distinguish. This is my real world life, and this is the online life. And in my real world life, there are very few people who I trust fully and who I think I love and love me. My parents will always be on my side, no matter what crazy shit I do online. And my wife and kids, they will always be on my side, no matter what crazy shit I do online.

And I think it is very, very important to distinguish. This is my real life family and friends, and they love me unconditionally, and I love them. And everything online is just noise. Even if everyone online or everyone outside this box hates me, it doesn’t matter. Because the most important people are in this box and the people that I love and love me back.

From Interview Coder to Cluely: the UX we didn’t know we needed

Interview Coder is a tool designed to let you cheat on technical interviews. But what we realized as we built Interview Coder is that the idea of an AI that sees your screen and hears your audio sort of shows itself as this translucent screen overlay. This has never really been attempted before. This is a completely novel user experience and it’s very short sighted to think that this is only good for cheating.

Ultimately what we’re building is we’re building for a future where models are multimodal and the models are not there yet. And they’re probably not going to be there yet for another three years. Nobody’s really thought of what happens when chatbots are no longer relevant. What happens when you don’t want to prompt GPT anymore and AI just knows what you want? Then how will you interact with AI? Nobody’s really attempted this before and I think Interview Coder was the first proof of concept of a user experience that could work in this world.

So we realized that and that’s, that’s what we’re building. Cluely off. I mean Cluely is the new way you will use AI in, in five years, hopefully if we do things right, then in two years.

“Cheat on everything” - what that phrase actually means

The phrase cheat on everything is intentionally ambiguous. Like what is cheat on? Like I know what cheat on test means, but I don’t know what cheat on everything means. It’s left to be sort of confusing and make you sit on it and reflect for a moment.

When you see someone using AI for everything, it makes you think this is unfair. They’re not supposed to be doing that, they’re cheating. In reality, like if you can use this for everything, like what does cheating on a meeting look like? That’s not really a thing. It’s just our gut human reaction to think this is so different, this is such a big advantage that it’s unfair.

And what we hope to do is we hope to give everyone this advantage. When every single person is using AI to cheat on meetings, then it’s not that you’re using, you’re cheating anymore. You’re just. This is just how humans will operate and think in the future.

I think when you can use AI, you should use AI. If it helps you, then you should use it. If using a calculator will help you, then you should use it. If using spellcheck will help you, then you will use it. Eventually the spellcheck will teach you how to spell the right words because you’ll get used to it so much or you just won’t need to know how to spell anymore, you’ll just need to know what the word is. If you can use AI to help, then you should. And if it can already do the job, then you’ll never need to do the job in the future.

Interviews are going to change. Maybe interviews go away

Assuming AI is everywhere, which it will be, all of technical interviews need to change. Not just in software engineering, but everywhere in the world. If you get asked a question and AI can answer the question, then you should probably get that out of your interview.

I think interviews will be a lot more holistic and I really question whether we even need job interviews at all in the future. If there is an AI that knows everything about you, everything you’re good at, why do you need a one hour interview to assess anything other than culture fit? I already know all the work you’ve done, or at least the AI already knows the work you’ve done. It knows how good it is, it knows what skills you’re good at. And if there is a skill match, then AI should just be able to match you directly to the job. Assuming that we get along after like a 30 minute conversation, I really don’t know that there is a need for interviews in today’s age.

But, but right now what we use is, it’s really just a conversation. We check if you’re a culture fit, we talk about past work you’ve done, and that’s pretty much it.

Cluely’s mission in one sentence

The whole point of Cluely is to get everybody used to the fact or used to a life where they use AI for everything. Once everybody uses AI in every instance possible, there’s going to be a lot of jobs that get replaced and there’s going to be a lot of people who are able to do so much more than they previously were.

If every scientist decided one day like today, I’m going to start using AI as much as possible too, they’ll be 100 times more productive. When scientists are 100 times more productive, we cure cancer 10 years earlier, we cure Alzheimer’s 10 years earlier, everyone lives to 400 years old and we’re on the next flight to fucking Mars in like two years, the rate of societal progression will just expand and exponentiate significantly. Once everyone gets along to the fact that we’re all using AI now, and that’s what Cluely hopes to achieve, is to get everybody used to, we’re all using AI now.

Building the overlay: more taste than raw tech

I think the, the user experience. I spent a lot of time making the user experience very seamless. It’s less of a technical challenge, I think and more of a taste challenge. The, the concept of a translucent screen overlay, something that really has never been attempted before. And it’s something that, that I tried and I think I only got to it after like dozens of iterations of different tools. That would be a more seamless use of AI in your life.

Yeah, I think that was probably the biggest technical challenge was just figuring out like what exactly is the best user experience for someone using this tool.

Latency, accuracy, and getting to instant

I mean, I mean latency, response speed and accuracy are like the two biggest things. This is what every model, every like OpenAI is working to improve latency and accuracy. There’s ways that we can get to a much faster response.

• If we host models on our own servers, this eliminates a lot of the latency that comes from the load balancing and request handling that is just inherent in OpenAI servers. That’s probably what we will end up doing.
• There’s ways that we can cache the input and sort of like parameterize the input so that you get the same condensed info, same information, but just condensed in a smaller way. And the smaller the input size, the faster the time to first token.
• Also generally accuracy can be improved by specific system prompts.

We’re developing custom evals in house based on a lot of the analytics and usage that we’re seeing and like everything is getting better. Like every single day the mod of the cluely gets more accurate and faster. At a certain point we’re going to know exactly what type of responses you prefer as an individual, what sort of conversations that you’re in and we can use all that data to generate a very, very hyper specific, personalized, fine tuned model for you that knows that hey, I’m a media reporter, I conduct these sorts of interviews and I generally want these types of responses or the tonality of my emails is this. So I would like you to respond in this way and you can just get the most personalized model in the world.

And once we have that data as like a moat defending us from the other big tech companies, then we’ll pretty much be unstoppable, more so than the data. I think the user experience is just interesting, untapped and novel. If we’re correct about this, then we’ll be the first to market and there’s a huge first mover advantage. When you’re trying a new form of UX and as if we can capture the market quickly enough by going viral enough sufficiently then I think there will be, it will be very hard to compete with us.

AI will change how humans think - not just how we work

The entire way we’re going to think will be changed. Every single one of my thoughts is formulated by the information I have at this moment. What happens when that information I have isn’t just what’s in my brain, but it’s everything that humanity has ever collected and put online, ever. What happens when AI literally helps me think in real time?

The entire way that humans will interact with each other, with the world, all of our thoughts will be changed. Like, what happens when I know about every single post you’ve made online ever? And I use that to distill down into like a, like a condensed blurb of everything about you, ever? What does our interaction look like then? It’s really hard to say, but I think this is a turning point for humanity and it will fundamentally change the way that we think and the way that we behave as humans.

If you’re not building in AI right now, what are you doing?

Well, if you’re not building a company in AI right now, then you’re probably not doing the right thing. AI just enables you to build such cool stuff and it’s such a new technology that even if you’re 19 and you’ve been playing with it for two months, you are one of the brightest minds. You are one of the pioneers of the field.

It’s not like biology, where if you haven’t studied for 10, 20 years and you don’t, then you’re not an expert in biology. You can study AI for two months and you’ll be an expert in AI. This technology is so gigantic and it’s so new that you can be really, really young and you can, you can know it more deeply than anyone else, and you’ll have the opportunity to build like a billion, $10 billion company out of it.

My only advice: take bigger risks

I would say take bigger risks. This is the only advice I have for anyone, really. You are smart enough, you’re capable enough, you’re hardworking enough. Just take bigger risks. If you take bigger risks and force yourself into positions where you have to make it, you’ll find that you’re a lot more hardworking than you thought you were. And you’ll also find that life gets a lot more interesting.

And very often the downside of risk is much smaller than you think. And the upside of risk is much bigger than you think. I wasn’t really like this five months ago. I mean, literally, like, like half a year ago, I was thinking I would just want to get a job at a big tech company. That’s all I want to do. And it wasn’t until very recently that I thought, like, oh, I actually want to build companies and go all in on this.

How risks stack - from tiny to “post the confidential PDF”

Taking risks, initially it means being willing to get rid of the constructs and limiting beliefs in your mind that make you think, hey, when I graduate, I have to be an engineer, lawyer, doctor or whatever. Just being willing to see what would happen if you didn’t do it. Every risk starts very small right now.

• The first risk I took was, hey, what if I. What if I built this tool and told nobody about it?
• Then the next risk I took was, what if I posted online but made it free and didn’t really associate with it so that people, more people would see it?
• And eventually the risk just snowballed and snowballed. Until now I’m like fully posting whatever confidential document Columbia University gives me because, like, I don’t care. Like, the risk is not. It’s not that much of a risk anymore. And I’ve grown used to it.

So grow used to taking bigger risks.

Privilege, perspective, and why I don’t play the victim

I feel like my life is very easy. My life has been very easy. I mean, my mom, I’ve got loving parents. My mom, she made me study even when I didn’t want to. She would just say, well, you should go study. So I studied and as a result, I did well in school and I hung out with smart kids and they helped me do better in life. Like, I have two amazing, great parents and I come from a great family. I don’t really feel like my life has been all that challenging.

Getting kicked out of Columbia University, it’s not that challenging when you’re out there building companies and I was going to drop out anyways, getting rescinded from Harvard. This is also not that challenging when you have a loving family at home. There’s kids like out there who are starving in like Uganda. And like, my life is not that hard really. A lot of everything that you should think is you should just try and think more positively about life and be more optimistic about things.

It is very rare that you’re going to be in America, you’re going to have the opportunity to go to college and you’re really in like an actually challenging situation. In reality, we are in the most interesting time in history. If you live in America and you’re not like in poverty and your parents aren’t crackheads, you have the opportunity to make billions of dollars and make generational wealth and do the most interesting thing ever. There’s very few challenging situations that are so challenging that you’re just like limited right now. Anybody can do anything. And you should just try and take risks and be bold because you’re very privileged right now to be living in this world.

What success looks like to me

Success is having a wife, having 12 kids, and having people remember me. I think Steve Jobs and Elon Musk are very cool in that everyone has a strong opinion about them, whether it’s good or bad. Everyone has something to say about Elon Musk and I think that’s really cool.

We’re all going to die eventually and nobody’s going to remember us in a thousand years. I might as well be remembered as strongly as possible for the time that I’m here.

The real unlock: confidence

I think the biggest thing is confidence. Like truly, you hear all the time that the people that build big companies are not geniuses. They’re not smarter than you. They’re just, they take more risks than you and they’re hard work, harder workers than you. And I think this is generally true.

Five months ago I was just some random student at some random school and I didn’t really have anything going for me. And now I just raised $5 million and I’m in this giant office and I’m building a company that I hope will change the world one day. And very little has changed about me except the fact that I took a risk.

Even moving forward, if I do end up becoming like the next trillionaire, like as big as Mark Zuckerberg, there will be nothing about me that changed. It’ll just be a series of well calculated risks that I took that will lead me there. And I think the gap between Mark Zuckerberg and your average human, it’s really not that big. And if you just have the confidence to take bigger risks, then very often you will win.

Where we are right now

We launched Cluely - Cluely - about a month ago. We’re closing in on $5 million in ARR. We just closed a $5.3 million seed round led by Abstract Ventures and Suse Ventures. Those numbers matter, but not because they’re trophies. They matter because they buy time. Time to turn a viral hack into a durable product. Time to learn what the overlay needs to feel like in a world where AI will see, hear, and think alongside you. Time to make “cheat on everything” stop sounding dirty and start sounding normal.