GameChanger Innovation, Technology & Agility

What is the unique story that you feel your company has with AI? If you were writing about it, what would the title of your blog be?
Our early work in downtime detection focused on recognizing the rhythm of youth sports, pinpointing when play starts, pauses and shifts. These weren’t polished, high-budget broadcasts. They were handheld cameras, unpredictable framing and all the beautiful chaos that comes with youth games.

As one of the earliest hires on the computer vision team, I helped build the models and infrastructure that made this possible. That foundation challenged us to move beyond traditional rules and build systems that could adapt to real-world conditions.

Today, we’re taking that even further. Our action recognition work now focuses on helping AI interpret how the game unfolds, not just when. It’s about watching the game like a coach, not a camera.

What are you most excited about in the field of AI right now?
A major milestone came when our downtime detection models could reliably trim full-length games down to only the moments that matter, at scale and in the wild. That meant our AI wasn’t just working; it was interpreting.

I was directly responsible for building the model architecture and training pipeline that made this possible. After months of iteration and experimentation, seeing our system accurately cut out downtime across thousands of videos felt like watching the model understand the game for the first time.

Now, what excites me most is pushing our action recognition even further, moving toward systems that grasp the tempo, intensity and tactical nuance of each play across different sports. What we’ve built so far is just the beginning. There’s a whole wave of AI-driven features we’re preparing to roll out.

How do you learn from one another and collaborate?
One of the greatest challenges we faced was the sheer variability of youth sports video — shaky footage, inconsistent angles and unpredictable pacing. Our downtime detection and action recognition models had to adapt to all of it.

What started as scrappy prototyping evolved into robust, production-scale systems, thanks to close collaboration between CV, engineering and design. We still operate with a startup mindset: fast iteration, open feedback loops and constant refinement.

We treat our models like living products, constantly learning and evolving based on real-world data. Continuous learning isn’t just about staying current with research — it’s about staying close to our users, our data and each other.