Computer Vision Engineer

eShepherd is virtual fencing for cattle: solar-powered neckbands, a phone app and a platform that lets a farmer draw a paddock on a screen and have the mob standing in it by morning.

We run on farms in more than twenty countries, from the Burdekin in Queensland to South Dakota prairie, from Texas brush country to Hawkes Bay hill country. We have gone from startup to scale up inside Gallagher, the New Zealand company that has been making farm fencing for 88 years. We move at startup pace with the reach and patience of a business that has seen a few cycles.

The Opportunity

Vision Weigh is the animal measurement product inside the eShepherd Farm Operating System™, sitting alongside virtual fencing and pasture intelligence. Cameras mounted in the paddock read the animal as it walks past, giving producers information they would normally only get by yarding and weighing. That data lands in the same platform that already holds the animal’s location, movement and grazing history. It is one of the hardest technical problems in our portfolio — and one of the most valuable to the farmers who need it.

You will take Vision Weigh from working prototype to hardware running reliably on commercial farms. This covers the model, the pipeline, the edge compute, the calibration, the ground truth data and the retraining loop that keeps the model accurate as deployment grows.

You will own the vision work, backed by specialists across data, software, firmware, hardware and product. The cycle here is design, test, deploy, iterate. We would rather have something useful running in a paddock this month and improve it every week than spend a year polishing something for a launch that may never happen.

The Fit

You turn ideas into working products. You are the engineer who would rather build the rough version, put it in front of a real animal, find out what breaks, and go again.

Progress over perfection – test it, learn quickly, make it better.

You are comfortable with ambiguity, be a subject matter expert in computer vision, make your own calls, put mistakes on the table early, and speak up when you are stuck.

You are curious about the physical world. The interesting problems in this role happen outdoors, where animals stand at awkward angles, lenses get covered in mud, low sun ruins a frame, breeds vary enormously and two animals walk through together.

We are office first, because the best ideas come out of working closely with the people around you.

What You’ll Bring

We’re looking for someone who knows how to take computer vision beyond the experiment and make it work in the real world.

• You have built computer vision into a real product, with something you wrote running somewhere beyond your own laptop.

• Strong Python and PyTorch skills, backed by a solid understanding of classical computer vision. Segmentation, keypoints, pose estimation and basic point cloud work should all feel familiar, and you know when geometry is the right tool for the job.

• Experience deploying models on edge hardware, whether that is a Jetson board, a camera module with an onboard NPU, or a single-board computer with an accelerator. You understand the realities of latency, memory and thermal limits.

• Experience with depth sensing, whether stereo, time of flight or structured light. You are comfortable with calibration and understand the challenges that come with noisy depth data.

• A pragmatic approach to data. You understand that a well built ground truth set is worth more than a clever architecture, and you have used experiment tracking and some form of dataset versioning so you know which model saw which data.

• Solid engineering habits. Version control, containers, continuous integration and enough testing that someone else can pick up your code and get it running.

• Real fluency with generative AI, in both directions. You use agentic coding tools to move faster and understand where vision foundation models genuinely add value — from reducing annotation effort to detection and pretrained depth models. We value engineers who can get strong results from a few thousand labelled animals rather than waiting for a hundred thousand.

Experience with depth SDKs such as Luxonis DepthAI, Stereolabs ZED or Intel RealSense, with AWS, Databricks or MLflow, with C++ on latency sensitive paths, or with livestock and agriculture, would all be highly regarded.

Why eShepherd

What we ship changes how farmers live. A bull breeder in Western Australia whose cows and calves have never done better. A grazier moving cattle on his Taranaki river flat from a deck chair in Wanaka. A Texas rancher who has freed up his labour and manages country more intensively than he ever has.

Animal welfare is the reason this product exists. Weighing and condition scoring normally means mustering, yarding and putting animals through a crush, which costs the producer time and costs the animal a stressful day. Getting the same numbers from a camera as the animal walks past to water removes all of that, and the accuracy bar we hold ourselves to is what makes it possible.

You would be joining a team that is number 8 wired, where resourcefulness beats resources and the default question is whether we can rather than why we cannot. Feedback runs in both directions, disagreements happen in the open, and the pace is real.

We also offer

• Competitive salary and performance-based incentives.

• The excitement of a high-growth startup, but with Gallagher stability.

• Career growth opportunities as we scale globally.

• Tools and tech that make your job easier - what you need to succeed.

• Supportive team culture – we value collaboration and practical problem-solving.

• KiwiSaver contributions to support your future.

• Boost App membership and staff discounts across NZ.

• Life & Disability Insurance – because looking after you and your loved ones matters.

• Hamilton site free parking, subsidised on-site café, and great eateries nearby.

• Employee Assistance Program and wellbeing initiatives.

We’ll be reviewing applications as they come through and may appoint before the closing date.