Gabriel Petersson
Research Scientist at OpenAI
How I Got Here
San Francisco, California. Gabriel Petersson is a research scientist at OpenAI, working on cutting-edge video generation research. His journey began when he dropped out of high school in Sweden at 17 to join the founding team at Depict.ai, a Y Combinator-backed startup. Since then, he has worked at Dataland (YC) and Midjourney.
Early Projects and Minecraft?
At the age of 17, Gabriel made the decision to leave high school to join Depict.ai, as a founding team member. In this role, he worked on building a product recommender system using computer vision (CNN) and natural language processing (NLP), learning on the job without formal training in engineering. For fun, at 14, he bought and sold Pokémon cards for over $20,000. At 18, he built a Minecraft server, which he subsequently monetized with an open-world RPG. Gabriel also gained millions of views on YouTube with Minecraft timelapses and led a Clash of Clans clan that was one of the first to hit 100 war wins.
Building High-Performance Software
At 19, Gabriel was recruited to become the interim CTO of Curb Food, the largest cloud kitchen in Sweden at the time, with 80 employees. In this capacity, he built a team of 7 engineers from the ground up, developing the tech systems necessary to scale the business. He successfully launched a kitchen management system into production, which streamlined operations and improved efficiency. Prior to this, at 18, Gabriel created a price comparison site for hand sanitizer during the COVID-19 pandemic, generating $22,000 in revenue in the first week. By strategically using viral-engineered Facebook posts and purchasing accounts from friends to maximize impressions, he reached 5% of the Swedish population. He also built and owned the video-call product troubleshooter at Mavenoid, a platform similar to Google Meet but with enhanced features such as live painting via WebRTC, which improved the remote support experience for users.
After, Gabriel joined Dataland, another YC-backed startup, as a founding engineer. He said, “I built one of the most performant web tables in production, capable of handling hundreds of thousands of rows, filters, and sorting without losing frames.” Next, Gabriel worked at Midjourney as a software engineer. There, he built the world’s most performant web image grid, designing a custom multithreaded rendering queue and using offscreen canvas techniques to smoothly render thousands of images. Gabriel also contributed by building internal tools for hyperparameter tuning, dataset exploration, and psychometrics.
Sora Research at OpenAI
Eventually, Gabriel shifted his focus from product engineering to AI research. At OpenAI, he works as a research scientist at Sora, developing video generation systems. Gabriel has previously written about Soras potential on X:, “AI still has a world understanding problem. But it’s possible to create worlds for AIs to explore, rediscover physics from raw pixels, and create simulations for humanoids to learn.”
The Road to the U.S. and the O-1 Visa
Gabriel’s path to the U.S. wasn’t exactly straightforward. Initially, he set his sights on the H-1B visa, only to realize it required a college degree—a barrier for someone with no formal education. Not one to be easily discouraged, Gabriel reached out to Plymouth for advice. That’s when he discovered he might actually be eligible for the O-1 visa, designed for individuals with extraordinary abilities.
I used Stack Overflow answers for the academic publishing criteria to get my O-1 visa. If you want to get into the U.S., get good lawyers who understand what matters for the government & how to stand out," Regarding Plymouth, Gabriel said. “They asked all the right questions and helped me identify paths I never would have thought of—things that counted towards the O-1 criteria."
Closing Advice
Don’t just follow what others say—think for yourself. The incentives around you are often misaligned, so it’s important to step back and find your own way. True growth comes from questioning the status quo and trusting your own judgment.