We're looking for a data scientist to own computer vision as we open up a new product area.

The work is image analysis at scale: finding the individual objects that matter in overhead imagery, classifying their physical condition, and tracking how that condition changes over time. The output has to be something a professional will act on, which means every grade needs evidence attached and an honest confidence attached to it.

We're strong on remote sensing and statistical modelling. Nobody here is a computer vision specialist. That's the gap you'd fill.

What you'll own

• Turn large volumes of raw imagery into clean, isolated objects — geometry, alignment, occlusion, the unglamorous parts

• Build and validate the visual features that actually separate one condition class from another

• Train and evaluate models on small, expert-labelled datasets, and be straight about what the numbers prove

• Compare the same object across time, so a single snapshot becomes a trajectory

• Design the labelling workflow together with the domain experts who define what the classes mean

• Turn model output into evidence a non-specialist will look at and trust

You'll work alongside our Head of Data, who owns remote sensing and satellite modelling, and our Lead Risk Modeller, who owns the statistical and risk side. On computer vision, you'll be the owner.

This is a first or second job, and we're hiring for judgment and craft rather than years. You'll have finished an MSc or PhD in computer science, applied mathematics, physics or similar, and you should be genuinely strong in computer vision — segmentation, detection, and the classical techniques like texture and morphology, not only fine-tuning something pretrained. You do not need a remote sensing background; we have that and we'll teach you what you need.

We'd rather be honest about the hard part: the datasets are small, the labels are inconsistent, and a real share of the work is deciding what the model genuinely cannot tell you. If that sounds like the interesting part rather than the frustrating one, we should talk.

Danish isn't required. We work in English.

About Envira

Envira turns Earth observation data and AI analytics into finance-ready intelligence about climate and environmental risk. Traditional datasets look backwards. Our customers need to look forward — how exposed is this address to flooding, how will this season affect this field, what does this portfolio look like under climate stress.

Insurers, banks, agribusinesses and public agencies use our products to answer those questions. We won the EU Big Data from Space Award 2025, are a finalist for the EARSC European Start-Up Award 2026, and are backed by NextGenerationEU and Danish innovation funding.

We're a small team in Denmark. That's the reason this role has the scope it does.

How we build

AI-assisted development is the basis of all technical work here, not an optional extra. A large share of our codebase was written in collaboration with AI coding tools, and we expect the same from you from day one. This is a hard requirement, and it's the main reason a small team can take on work at this scope.

We're not asking whether you've tried an AI assistant. We're asking whether you can work at a high level with these tools — direct them precisely, read and review what comes back, throw away the parts that are wrong, and stay fully accountable for what you ship.

It raises the bar on judgment rather than lowering it. That's doubly true in modelling work, where it's trivially easy to generate a pipeline that runs, produces a number, and means nothing. The people who do well here use these tools to get to the simplest thing that actually answers the question, then stop.