A Note from the Founder

I started Train Matricx because of one stubborn problem: sports footage is easy to label badly.

A general annotation team can draw a box around a player. Ask the same team whether that was a tackle or a foul, which of two overlapping players has the ball, or where a cricket ball is in the three frames where it is a blur behind the bat, and the labels start to drift. The model trained on them inherits every one of those mistakes.

So we built the company around one rule. The people who label a sport should understand that sport. Our annotators include former athletes, coaches and statisticians. They know the rules, the roles and the camera angles, so they can tell what happened in a frame and not only where the pixels are.

Why people, when models can label?

I get asked this often, and it is a fair question. Automated labelling is fast and it keeps getting better.

But sport is made of the cases automation finds hardest: ten players in a goalmouth, a puck against white ice, a ball that leaves the frame and comes back. Those are the frames a model most needs to get right, and they are the ones where a person who knows the game still makes the better call. Every model that reads a match today learned from frames a person once labelled. We take that part seriously.

What we hold ourselves to

Every delivery passes through three layers of review: the annotator, a sport-specialist reviewer, and a senior auditor. Each one ships with a QA report you can check frame by frame. If a batch falls below the threshold we agreed with you, we re-annotate it at no cost.

We sign an NDA before any footage moves. We work to your schema and your format, not ours.

And we do not ask anyone to take this on trust. Send us your hardest clip. We will annotate it free and return it within two business days with the QA report. If it is not good enough, you have lost nothing.

Our work with Dark Horse AI shows what that looks like at scale. It has passed 6,000 youth football matches at 98% accepted first pass, and you can read the details in the case study.

Where this is going

Sports AI is moving quickly: player tracking, automated highlights, officiating support, scouting from a phone camera. All of it rests on labelled data, and the quality of that data sets the ceiling for everything built on top of it.

I want Train Matricx to be the team that sports AI companies trust with that foundation, across more than 20 sports today and more as the field grows.

If you are building something in this space, I would like to hear about it. My inbox is open.