Pranav Bhuvanagiri: Student App Developer

February, twenty twenty-one. A high school student in Sugar Land, Texas hits publish on an app called SAT Daily. Three questions a day, a word of the day, free, no ads. His name is on the App Store listing as the seller... because he IS the company.

How old are we talking?

Still at Clements High School. His diploma is dated twenty twenty-two. So he's shipping a national test-prep product while he's, you know, taking tests.

That's a very specific kind of revenge.

And here's the thing I want you to hold onto for the whole episode. That app claims an average eighty point score increase from its users. Big claim. We'll come back to whether that holds up.

Oh, I'm absolutely coming back to that.

The pitch in the listing is almost touchingly student-written. Quote, "designed by students that have attained a score of fifteen hundred and higher." Use it waiting for your bus. Use it standing in line.

Which is exactly where you'd use it. Nobody sits down for a noble ninety-minute study block.

Right. And the build pace is the real story. SAT Daily launches in February. By July fifteenth, twenty twenty-one, he's pushed an Android version to GitHub, written in Java. And four days before THAT, July eleventh, he starts a completely different app.

Wait, in the same summer?

Same summer. It's called EyeglassML. The idea: point your phone at an object or an animal, take a picture, and get back not just the name but a whole fact sheet. Chemical composition. History. Common names. Even, his words, "mythical stories about the object."

That's not a weekend project, that's a Google feature.

It was a weekend project. Two-day hackathon. Hack the Cloud. Forty-eight hours.

Okay, so how does one person build computer vision in two days?

He doesn't, really. And this is the bit I love, because he wrote it down honestly on Devpost. He TRIED to train his own machine learning model. Couldn't find the right datasets. Didn't have the computing power. So he gave up on it and used an off-the-shelf model, MobileNetV2, with Apple's Vision and CoreML.

Giving up as a strategy. I respect it.

Then the second wall. For the facts, he's pulling from Wikipedia through its API. He gets it working... and the volume of data coming back is overwhelming. He says filtering all of it inside forty-eight hours would eat the whole clock.

So what does he do?

He writes, and I'm quoting, "I settled on a hacky fix that filters only the important data and somewhat ignores the unnecessary data."

"Somewhat ignores." That's the most honest sentence in software.

And it placed fourth. A hacky fix, a borrowed model, shipped. He even says afterward the whole thing needs a complete UI rewrite because he built it in UIKit for speed when he's actually better at SwiftUI.

So he knowingly built it the worse way to make the deadline.

That's the lesson of the entire episode, and we're only halfway.

Alright, the eighty points. Is that real?

Here's what I can actually stand behind. The App Store listing says users attributed an average eighty point increase. That's the developer's own claim, not an independent study, and there's no published data behind it I could find. What IS verifiable: the project's site says over ten thousand downloads. And the app has eleven ratings, averaging three point four stars.

Eleven. Out of ten thousand.

Welcome to indie apps.

And three point four is... fine? That's a "it works but" rating.

It is. And notice the business model wobbles in public. The description says "absolutely no micro-transactions, we are a non-profit team." But the same listing shows a premium tier at nine ninety-nine.

Somebody discovered revenue.

Then came ACT Daily, same formula. Three daily questions, word of the day, score prediction. And he picks up paid work at a startup called Lockcard in July twenty twenty-one, which teaches you English vocabulary through phone notifications.

Which is the exact same insight as SAT Daily. Catch people in the dead seconds.

And here's where it lands. By twenty twenty-four he's a junior at UT Austin studying computer science and mathematics, interested in cloud, mobile and machine learning, and interning as a software engineer at Fidelity Investments. He'd also TA'd intro to programming and worked in a research lab there.

So the apps were the resume.

The apps were the resume. And in the fall of twenty twenty-four he's writing blog posts about grinding LeetCode and admitting, openly, that it's very easy to get demotivated doing it.

That's weirdly reassuring. The kid who shipped an app in high school still hates the interview.

Everybody does. The difference is what's in his hands when he walks in.

So what's the takeaway I can actually use?

Ship the hacky fix. His own words: he somewhat ignored the unnecessary data, borrowed somebody else's model, built it in the framework he liked less... and placed fourth with a working thing. The custom model he dreamed about? Never shipped. Nobody's grading your architecture at midnight.

They're grading whether it opens.

Three questions a day. Forty-eight hours. Ten thousand downloads. None of it was impressive on paper... until it existed.