The Calorie Counter Built in a Hacker House

At sixteen, Zach Yadegari sold a website for $100,000. It was called Totally Science, and it did one useful thing: it let students play games on school Chromebooks after the school had blocked the usual game sites. According to TechCrunch, he built it in ninth grade, named it to look like homework so that the filters would leave it alone, and sold it to a gaming company called FreezeNova. Fortune reports that the site reached about five million users.

That is a respectable start for a teenager and an odd one for a founder of a health app, but the first business supplied two things the second one needed: a little money, and proof that Yadegari could get strangers to show up.

Cal AI

The second business is Cal AI. The app asks you to photograph your meal and returns calories and macronutrients. Per TechCrunch, Yadegari co-founded it at 17, while a high school senior. It launched in May 2024, Fortune says, with co-founders Henry Langmack, Blake Anderson and Jake Castillo.

The product was not novel, and TechCrunch said so at the time: MyFitnessPal already had a meal-scanning feature, and an app called SnapCalorie, from the founder of Google Lens, had shipped in 2023. Under the hood, TechCrunch reported, Cal AI runs on models from Anthropic and OpenAI plus retrieval over open food databases. Yadegari’s own finding was that “different models are better with different foods,” so the app routes accordingly. The team claimed roughly 90 percent accuracy.

What made it a business was not the model. In Yadegari’s words to Fortune: “We started making TikToks, we started partnering with influencers, and then we had pretty explosive growth.” The company ran influencer marketing out of its own staff, led by Castillo, and the founders lived in a San Francisco hacker house for a stretch, which Yadegari described to TechCrunch as “twenty-four-seven grinding, sleeping on the floor.”

The numbers, and who said them

By March 2025 Yadegari told TechCrunch the app had passed five million downloads in eight months, with retention above 30 percent and more than $2 million in revenue the previous month. TechCrunch was explicit that it could not verify those figures. What it could see was public: a 4.8-star rating on the App Store and over a million Google Play downloads.

Six months later Fortune said it had reviewed financial records showing several million dollars in monthly revenue, and reported $30 million in annual revenue on a free-to-download app with subscriptions at $2.49 a month or $29.99 a year. A Business Insider as-told-to essay said the magazine had verified the financial claims in it. By September 2026, TechCrunch was writing, in a conference promotion that cites no sourcing, that Cal AI had passed 15 million downloads and $30 million in annual revenue, and that MyFitnessPal had acquired it in December 2025. No price was disclosed. The same page says the team kept was seven people; Fortune had put the staff above thirty a few months earlier. Treat headcount as a number that depends on who is counting and when.

The college post

One piece of the story had nothing to do with software. In April 2025 Yadegari posted an essay about being rejected by 15 of the 18 colleges he applied to, including Stanford, MIT and Harvard, despite a reported 4.0 GPA and 34 ACT (his figures). TechCrunch counted more than 22 million views on the post, and a lot of readers found it arrogant. He wrote that “ultimately, I’ll never know exactly why I was turned down.” He enrolled at the University of Miami, where Fortune found him, in his words, “paying $100,000 a year for a vacation.”

It would be tidy to say the rejection drove the company, and nothing reported supports that. What the episode does show is that a founder’s personal feed is itself distribution. A post that reaches 22 million people is a marketing asset no seed-stage company can buy, and it arrived on top of a product already being pushed by TikTok creators.

What this is evidence of

Three things, none of them flattering to the old path or to the new one.

First, the AI wrapper is a legitimate business when the wrapper is attached to distribution. Cal AI did not train a frontier model; it rented two, tuned the routing, and spent its energy on creators. The moat, if there is one, is the audience and the App Store reviews.

Second, building in public works as a recruiting and credibility engine. Anderson met Yadegari on X, according to TechCrunch, and the two high schoolers studied Y Combinator videos and networked with coders there.

Third, luck was substantial and the sources do not hide it. Totally Science existed because schools blocked game sites. Vision models became good enough for food photos at roughly the moment two teenagers wanted to build one. The crowded market, as TechCrunch noted, meant that being first was never available. That Cal AI won anyway says more about execution on marketing than about invention. It also leaves open how durable a calorie counter is once the incumbent it sold to bundles the same feature.

The exit was a sale to an incumbent that already had a competing feature, and it came, TechCrunch notes, before Yadegari’s twentieth birthday. The college question that once looked like his main problem became a footnote.