SeeFood: Hotdog or Not Hotdog

I always dreamed of working for a Silicon Valley tech company. I’ve visited Palo Alto a few times, seen the wider SV ecosystem firsthand, and realised: HBO’s Silicon Valley isn’t exaggerating. If anything, it’s generous. The absurdity is real. The hustle, the failed pivots, the obsession with being ahead of innovation—Erlich Bachmann isn’t fiction, he’s documentary.

I rewatch the show regularly now. I’ve introduced it to my wife and daughter too—it’s become a way to explain how Silicon Valley actually works, what the culture really is. It’s one of those shows that gets better when you’ve lived in the world it’s satirizing. You catch the stuff that seemed like exaggeration and realise it’s just observation. During a recent rewatch, I hit the episode where Jin Yang’s app launches: SeeFood. Shazam for food. A wildly impressive food recognition app, thrown together at a hackathon, everyone assumed it would be the next big thing.

Except it only did one thing: identify hotdogs.

The joke was perfect. A technically impressive app with zero market fit, deliberately constrained because of the absurd nature of how ideas actually develop in that world. Hotdog or not hotdog. That’s it.

Watching it this time, a thought landed: with modern AI, you could actually build this. Not as a joke for a TV show, but as a real app. And more importantly, you could lean into the joke—make something that genuinely solves the problem with unnecessary precision, on purpose.

The infrastructure existed now. Core ML. Vision framework. SwiftUI. The tooling had caught up to the parody.

Building It

I named it SeeFood by Aviato. If you’ve watched the show, you know why—Erlich spends the entire series scheming and hustling to own the next big thing, convinced he needs to be clever and strategic to get ahead of innovation. He’s obsessed with buying the SeeFood domain from Jin Yang. That was actually part of the inspiration.

Meanwhile, you just build it. No grand scheme. No narrative ownership. Just: the infrastructure is good enough now, so why not.

The app itself is technically overkill for what it does. You’re using computer vision to solve a question that matters to approximately zero people. But that’s the point. The app doesn’t apologise. It launches with confidence. You point your camera at something. It tells you if it’s a hotdog or not.

Everything is hotdog or not hotdog. There’s no middle ground. No “maybe hotdog.” No “hotdog-adjacent.” It’s binary. Absolute. Occasionally wrong, which adds to the comedy.

A real food app would scan ingredients, identify quantities, establish macro balance, calculate protein levels. It would be useful. Instead, this app looks at your dinner and answers one question with unnecessary conviction. That’s the whole point. Willful ignorance wrapped in confidence.

The implementation was straightforward. Core ML handles the model. The Vision framework handles the camera and real-time inference. SwiftUI handles the interface.

The Joke Still Works

What’s interesting is that the joke doesn’t wear thin. You launch the app expecting it to be funny for thirty seconds. Instead, you find yourself testing it. You point it at things that are definitely hotdogs to see if it agrees. You point it at things that are vaguely hotdog-shaped to see where the boundary is.

The app has opinions. Apparently, some sausages aren’t hotdogs. Some bread-adjacent things are definitely hotdogs. The model has learned something about what constitutes a hotdog that isn’t just “sausage in bread.”

Building a Parody in Earnest

There’s a specific type of project where the joke is the entire product, but you build it with complete seriousness. SeeFood is that. It’s not winking at the audience. It’s not trying to be funny. It’s trying to answer one question with maximum precision and zero compromise.

That’s the sweet spot between parody and genuine product. You’re making fun of something (AI apps that do one thing well, but that one thing is useless). But you’re also making something that actually works.

The Result

SeeFood works. Point your phone at something. It tells you if it’s a hotdog or not hotdog. The confidence score is right there. Most of the time it’s right. Sometimes it’s confidently wrong, which is part of the appeal.

There’s something deeply satisfying about an app that knows exactly what it does and refuses to do anything else. No pivot to “other foods.” No feature creep. Just the one question, answered with maximum conviction.

The joke was always that someone would build this eventually. The interesting part is that when the moment came, it turned out to be trivial. A rewatch, a thought, an afternoon of code. The infrastructure had moved so far that impossibility became routine.

That matters. Not for this specific app—who cares about a hotdog classifier. But for what it says about what’s possible now. Build things because they’re fun. Build things because they’re jokes. The technology will get out of your way.

If you want to test it yourself, SeeFood by Aviato is available on the App Store.