Halter

Devon Copp

Machine Learning Engineer

ZOZO NZ
Auckland, New ZealandProjects devoncopp@gmail.comLinkedIn

I own machine learning and AI projects end to end, from framing the problem to shipping the system around the model.

Models
Designed and trained for the problem, from vision to graph neural networks.
Systems
Deployment, MLOps and agent tooling, on device or in the cloud.
Product
Built for the people who use it, with care for the design.
Devon Copp smiling at night in a city square, a lit statue behind him.

Selected work

Halter

2021 to 2023

Smart collars that let farmers fence, move and watch their herd from an app. I joined as employee number 64; Halter is now a $2 billion company, and I’m proud of what I built there.

Dairy cows grazing a wide paddock, with a clear line across the grass between grazed and ungrazed pasture.

Algorithms for cow safety

A tree falls across the raceway. The first collars to reach it sense trouble and tell the rest over Bluetooth. The whole herd locks out, and no cow gets a pulse she doesn’t need.

Problem
An obstruction can trap a herd against cues it can't follow.
My part
The detection algorithm: spotting the obstruction from the first few cows to reach it.
Result
A patented herd-wide lockout, built with a small team of engineers.

Turn lockout off to see what it prevents. A toy herd, not Halter’s algorithm.

Power for a 24/7 wearable

The collar is small, solar powered and never comes off, so every milliwatt counts.

Problem
Run GPS, radio, sound and sensing all day and night on sunlight.
My part
Behaviour detection and power algorithms in the firmware, built on lightweight signal processing.
Result
Components switch off when they aren't needed, with full function and safety kept.
GPS fixes, last six hoursMotion sensor, GPS on-time beneathBattery with GPS always onBattery with motion-gated GPS

0%of the toy cow’s day with GPS awake. Now 06:00, lying down, GPS asleep.

One example: GPS wakes only when she moves. A toy model, not Halter data.

ZOZO New Zealand

2023 to now

The Auckland team behind ZOZOFIT, building measurement technology for ZOZO in Japan. As a machine learning engineer I own models from data and training through to the phone.

Body scanning on the phone

ZOZOFIT measures your body from photos taken on your own phone.

Problem
A scan has to be quick and dependable on whatever phone you own.
My part
Trained and deployed the deep learning vision models behind the scan.
Result
A faster, steadier scan, in production.
My own ZOZOFIT body scan, redrawn in ink: a full-length 3D body model covered in its wireframe grid.

What I work on

Computer vision
Deep learning models that run on the phone.
Recommendations
Systems that match people with what suits them.
MLOps
Training, evaluation and deployment pipelines the team relies on.
Agentic AI
Agents and tools that take repetitive work off people’s plates.

How I work

I own the whole life cycle, from the first conversation to a system people use. I care about the product, and I enjoy the design side.

  1. Frame

    Find the real problem with the people who have it.

  2. Measure

    Build the yardstick before the model.

  3. Iterate

    Change one thing, measure, keep what wins.

  4. Ship

    Deploy, monitor and hand over a system that stays good.

Need a model built, or an existing one wired into your product? I’d like to hear about it.

Have a project in mind?

devoncopp@gmail.comLinkedIn

Auckland, New Zealand