—
In 2020, most of the world was not thinking about artificial intelligence. The conversations dominating tech were about remote work, supply chains, and whether Zoom could handle the load. Large language models had not yet captured the public imagination. The generative AI boom was still years away.
Erhan Ciris was thinking about something else entirely. He was trying to teach machines to see.
Not see in the way a camera sees, passively recording whatever passes in front of the lens. Ciris wanted to build AI that could understand what it was looking at. Geometry. Depth. Motion. Lighting. The full spatial structure of a live environment, interpreted and acted on in real time.
That year, Ciris and his team began the deep learning research that would eventually become the foundation of 4D Sight, the New York City-based technology company he leads as founder and CEO. At the time, there was no playbook for what they were building. Real-time perception AI applied to live video was not an established category. There were no incumbents to study and no market consensus that the problem was even worth solving.
Ciris decided it was.
Solving the Problem Before the Industry Knew It Existed
The challenge Erhan Ciris was focused on in 2020 was specific and technical. Live video, particularly in sports and esports, moves fast. Camera angles shift. Lighting changes. Athletes and players move unpredictably. To insert any kind of digital content into that environment and have it look real, an AI system has to understand the three-dimensional structure of the scene as it happens, not after the fact.
Most computer vision research at the time was oriented around post-production or static image analysis. Real-time spatial reasoning inside a live broadcast, with no latency tolerance and no room for visible error, was a different problem altogether.
Ciris assembled a team with backgrounds in autonomous robotics, UAV navigation, and synthetic aperture radar imaging. These were engineers who had spent their careers building systems that had to perceive and react to physical environments under pressure. The transition to live media was a new application, but the underlying discipline was the same.
“We were not chasing a trend,” Ciris has said. “We were solving a problem that required years of foundational research before it could work in production.”
That research led to U.S. Patent 11,270,517, which protects the core method 4D Sight uses to dynamically insert content into live video streams.
What the Head Start Produced
By the time the broader AI conversation exploded in 2022 and 2023, 4D Sight had already been refining its perception models for years. While other companies were racing to apply newly available AI tools to media and advertising, Ciris and his team had a working platform that had been tested under real broadcast conditions.
4D Sight’s system processes a single live video feed in the cloud, interprets the spatial structure of the broadcast environment, and inserts photorealistic virtual content that behaves as though it physically exists in the scene. No on-site hardware. No manual calibration. Latency measured in milliseconds.
The platform was first validated in esports, where Erhan Ciris deliberately chose to start because of the format’s extreme speed and visual complexity. Partnerships with organizations like Riot Games proved the technology could perform under the most demanding conditions in live media.
In 2023, 4D Sight expanded into traditional sports through a multi-year partnership with TKO, which includes the UFC and WWE. The move confirmed that the same perception engine built for the chaos of competitive gaming could operate with broadcast-grade precision inside physical arenas.
None of that would have been possible without the years of research Ciris committed to before anyone was paying attention.
Why Starting Early Changed Everything
There is a difference between companies that adopt AI and companies that build on years of original AI research. Erhan Ciris and 4D Sight sit firmly on the research side of that line.
The deep learning models powering 4D Sight were not assembled from off-the-shelf tools or adapted from general-purpose frameworks. They were purpose-built for a single, hard problem: understanding live three-dimensional environments in real time with zero tolerance for failure.
That specificity is what gives 4D Sight its edge. In live broadcasting, there is no retry. A graphic that flickers, a virtual asset that drifts out of alignment, a system that introduces even a moment of visible latency, all of it is immediately apparent to millions of viewers. The margin for error is essentially zero.
Ciris built 4D Sight to operate inside that margin. And the reason the company can do it today is because the research started in 2020, when the rest of the industry was looking somewhere else.
The AI landscape has changed dramatically since then. Hundreds of companies now claim real-time capabilities. But for Erhan Ciris and 4D Sight, real-time perception is not a recent addition to the product roadmap. It is the foundation the entire company was built on.
—
