Machine Learning Engineer
About the role
Cub Digital does the ordinary things. They just do them extraordinarily.
Most agencies build dashboards. Cub builds the thing that watches an All Blacks training session and tells the coach what the human eye missed.
Most teams ship a model and call it done. Cub ships it into production, watches it run at 200+ frames per second, and asks what's next.That's the bar. Here's who clears it.
You've trained and deployed computer vision in production, not in a notebook. Panoptic segmentation, object detection, pose estimation built for real-world speed, not demo-day applause. You know the difference between a model that works and a model that survives contact with production traffic.
You know when to build and when to borrow. YOLOv8, SAM, the foundation models that get you 80% of the way there in a fraction of the time...you use them like tools, not crutches, and you know exactly where the remaining 20% lives.
You think across the stack, not just inside the model. PyTorch through to Docker, ONNX, TensorRT, you don't hand your model to someone else and hope. You own it end to end.
You've done the unglamorous data work that makes the glamorous model possible. Mining visual-language data at scale, finding the rare scenes nobody thought to label, building the dataset that makes the model actually good instead of just technically functional.
You're curious in a way that doesn't switch off at 5pm. Personal projects, side research, the kind of person who reads the paper before someone tells them to.
If that's not a checklist but a mirror, if you read that and thought "yeah, that's just Tuesday" Cub Digital wants to talk to you.
This is a small, sharp team working on problems with actual stakes: elite sport, financial services, systems where "close enough" isn't a tier that exists. You'll be in the room with the people making the calls, not three layers removed from them.
Auckland CBD. In-office, five days because the best version of this work happens when you can turn your screen and show someone what you just built, not schedule a meeting about it.
Apply through Potentia. We read every CV like it might be the one that changes the search.