Senior Devops / MLops Engineer

Inviol is one of New Zealand's fastest-growing startups, on a mission to save lives by turning AI insights into safer worksites. We plug into the CCTV cameras a business already has, spot the risky moments no one was watching, and turn them into coaching instead of incident reports. Sites using us cut operational health and safety risk by more than 50% within weeks. Coca-Cola and NZ Post are among the names that trust us with it.

We're hiring a Senior DevOps / MLOps Engineer to own the infrastructure that makes all of it run.

Three moving parts, one owner

Our cloud platform is serverless — Azure and Cloudflare, deployed with Pulumi. Our second production environment is a fleet of NVIDIA Jetson devices bolted into warehouses, ports and cold stores around the world, running GPU inference on live camera feeds, behind our customers' firewalls, well out of arm's reach. And the third is the models themselves: computer vision that gets retrained, re-evaluated and re-shipped to every one of those devices on its own schedule.

Our product is the models. A better detection model is worth nothing until it's running reliably on every camera in a cold store on a remote industrial site — ensuring that is what this role exists to do.

What you'll actually do

• Own the infrastructure. Pulumi across Azure and Cloudflare — stacks, environments, drift. It's yours to shape.
• Make deploys boring. CI/CD across Azure DevOps and GitHub Actions, staged rollout, zero-downtime swaps, rollback that works when it's 4pm on a Friday.
• Run the fleet. Provisioning, imaging, multi-arch container builds, remote config and health monitoring for edge devices you'll never physically touch. Where the stack goes next is a call you'll help make.
• Ship the models. Own the road from trained model to production GPU — versioning, registry, evaluation gates, staged rollout across the fleet, and rollback when a model behaves worse in a real warehouse than it did on the benchmark.
• Build the ML plumbing. Training infrastructure, dataset versioning, annotation pipelines (CVAT), experiment reproducibility. Our Model Training team should be thinking about models, not YAML.
• Watch the models in the wild. Detection rates, false positive trends, per-site and per-camera drift — surfaced before a customer picks up the phone.
• See everything. Application Insights, OpenTelemetry, Logpush. Alerting that's worth waking up for, and nothing that isn't.
• Watch the spend. Cost per site is an engineering number, and you'll own it.

What's in it for you?

• Impact & purpose. Every improvement to uptime is a camera that keeps watching a forklift aisle. The work has a body count it prevents, not a vanity metric.
• Real ownership. You'll be a senior voice on a small team. No layers to route decisions through.
• Hard problems. Distributed inference on other people's networks is genuinely difficult, and it's ours to solve.
• Growth. Fast-growing NZ startup expanding into Australia and the US — the surface area grows faster than the team does.
• People-first culture. Trust and curiosity, direct access to the founder, and no appetite for process for its own sake.

Who we're looking for

• Experienced. 3+ years in DevOps, SRE or platform engineering, with real production ownership behind you.
• IaC-fluent. Pulumi, Terraform or equivalent — as an owner, not a user.
• Container-deep. Linux, Docker, multi-arch builds, and networking you can reason about under pressure.
• ML-literate. You've put models into production and owned what happens next. You don't need to train them — you need to make shipping them boring.
• Python-comfortable. Enough to work inside ML tooling and training code, not just around it.
• Pragmatic. You know the difference between the right architecture and the right architecture now.
• An automator. Manual process offends you slightly.

Bonus points for

Edge, IoT or fleet experience • NVIDIA Jetson, DeepStream, TensorRT or video pipelines (RTSP/ffmpeg) • computer vision and models • MLflow, Weights & Biases, DVC or similar • CVAT or other annotation tooling • running VLMs/LLMs in production • Kubernetes deep enough to make an architecture call • Cloudflare Workers/D1/KV/R2 • startup experience

Details

Auckland-based, hybrid, permanent full-time.

How to apply

Send your CV and a cover letter — preferably one you wrote yourself — telling us why inviol and why this role. If you've ever had to fix something running on a machine you couldn't reach, lead with that story.