He Sold Mangoes. Now He’s Trying to Archive How Humans Move.

Here’s a weird fucking founder story.

Raj Patel dropped out of Berkeley, used to sell mangoes, and now he’s building Human Archive, a company betting that the next great AI dataset won’t come from the internet.

Before AI, There Were Mangoes

Raj and his cousin Shloke were selling mangoes long before they were building AI infrastructure. At one point, they sold 16,000 of them and planted more than 100,000 trees.

Then came Berkeley. Then dropping out. Then Human Archive.

There’s something we like about that trajectory. It doesn't feel engineered for a pitch deck. It feels like two people who keep finding weird problems and deciding to go after them.

AI has basically swallowed the internet. Books, code, images, conversations, all of it. But the physical world is still mostly undocumented. A great mechanic doesn’t just know what to do. They know how much pressure to use, what feels wrong, when to stop, how to recover when shit goes sideways.

You can't scrape that from the internet.

So Human Archive is trying to capture it. Cameras, sensors, motion, touch, real people doing real work. The goal is to turn human behavior into training data for physical AI.

It sounds simple. It’s hard.

The Bet

If robots are going to become actual workers, someone needs to teach them what humans actually do.

Not what we say we do.

What we see, touch, move, fuck up and figure out on the fly.

Raj thinks that data is going to matter a lot.

We think the bet is fucking interesting.

The internet taught AI what humans say.

Human Archive wants to teach it what humans do.

That could end up being one hell of a dataset.

Budera Founder Files: interesting founders before they're obvious.