Jae Lee.

Before he was pitching at AWS re:Invent and speaking at World Economic Forum events, Jae Lee was a data scientist decoding threats for South Korea's Ministry of National Defense. He was good at it - award-winning, in fact. But somewhere between military cyber ops and Silicon Valley, he saw something nobody else wanted to touch: the fact that 80% of the world's data is video, and AI could barely watch a clip without losing the thread.

He co-founded TwelveLabs in 2021 to fix that. Not by wrapping existing models with a prettier API - but by building proprietary video foundation models from the ground up. It was, by his own admission, a painful journey. While competitors moved fast with shortcuts, TwelveLabs spent two and a half, three years doing the hard thing.

That patience looks like strategy now. TwelveLabs' flagship models - Marengo 2.7 and Pegasus 1.5 - can search, classify, and reason across video with what Lee describes as "human-level understanding." Marengo handles multimodal embeddings across 47 languages at 78.5% composite accuracy. Pegasus reasons continuously over up to two hours of video in a single pass. Together, they process over 10,000 hours of content per day at 60 times real-time speed.

The customers range from sports broadcasters to federal intelligence agencies. In-Q-Tel - the CIA's venture arm - is among TwelveLabs' investors, which tells you something about what serious video intelligence looks like in 2025.

Lee is a UC Berkeley computer science graduate who went through TechStars before landing in the video AI space. His path through Amazon and Samsung internships, then military data science, gave him something most founders lack: a pragmatic sense of what "deployment at scale" actually means before you've written a single fundraising deck.

He describes his founding conviction plainly: machines need video to understand the world the same way humans do. Text-only AI is, in his view, fundamentally incomplete. The camera is not just a data source - it is a language.

On the question of building versus wrapping: "It was a painful journey in the first like two and a half, three years because folks are flying by." But he held the line, and the models TwelveLabs built became the differentiation that $107 million in funding bought into.

Jae Lee's path to building one of the world's most-watched video AI companies runs through military intelligence, Amazon internships, and a willingness to spend three years on a bet most investors wouldn't touch.

His Korean roots are not background color. They are load-bearing. In January 2025, he joined SK Telecom's K-AI Alliance, serving on the board of the South Korea Foundation Model Association - working alongside Samsung, SK, and LG to ensure Korean AI competes at the global frontier.