Build AI for the physical world.

Galeio is a small, bootstrapped company working on AI for complex modalities, starting with Earth observation.

We're second-time founders and we started Galeio in Paris in 2024, after building our first company in the US. Now we're looking for a few more people to build it with us.

We deliver operational frontier AI for complex modalities, and we own the whole stack, from pretraining dedicated foundation models down to the fine-tuning that adapts them to the exact task a client cares about. The work is diverse, and we're looking for Swiss-army-knife people: comfortable explaining a model to a client one day, and digging into pretraining and adaptation strategies the next.

Our goal is to push modern AI to its limits on sensing data: the signals we use to observe the physical world, rather than the text and images everyone else is modelling. The questions that interest us there are still wide open, and most of them don't have textbook answers yet.

The data almost never behaves the way you'd like. It's shaped by physics, noise and imperfect measurements.

SAR is a good example of why we find it interesting: it can see the ground through clouds and in complete darkness, and yet it's very noisy and physics-driven. It's the kind of data that forces you to understand what you're looking at before you can hope to model it.

We follow the research closely and contribute back when we can, but what we really care about is seeing a good idea survive outside the paper, on real data and under real operational constraints. One month you might be training a new foundation model; the next, adapting it to a demanding deployment environment, building an operational system for a customer, or working through a technical problem with major institutional partners.

Because we run several projects at once, a question that comes up in one often ends up improving another, sometimes in a completely different domain. On a team this size nobody stays in a narrow role: you might spend a few days on model architecture, then move to data processing, distributed training, deployment, or understanding what a customer actually needs.

More than a particular profile, we're looking for a way of working.

The people we tend to click with like mathematics but find its assumptions a bit too clean for the real world, or enjoy building software but are tired of abstractions that never touch anything physical. They read a paper because they want to implement the idea, break it and see where it fails, and they dig into failures instead of working around them.

You don't need experience in Earth observation: we learned most of it on the job, and so will you. Strong fundamentals matter far more than a long list of keywords, so if you're comfortable with Python and PyTorch, at home on Linux, and have already built real things beyond coursework, you're probably closer to what we're looking for than you'd think.

We're looking for people with convictions and without FOMO: you should know what motivates you and not need a trend to tell you it matters.

You should also have a real relationship with AI. It's changing the way we work, how we relate to knowledge, and more. We want people who have thought seriously about where that goes. We don't need to agree on the strategy, but you should have genuine opinions about what comes next. If you've only just started using AI tools and haven't yet thought about how they change the way you work, this probably isn't the right moment for you to apply. We're not looking for a background, we're looking for character.

If self-supervised learning, distributed training or deep learning hold no mystery for you, even better.

Some of the most interesting data in the world is data almost nobody is looking at. If there's a kind of signal you've become obsessed with (hard to capture, harder to interpret, and nowhere near the hype cycle), tell us about it. Bonus points if you've built your own sensors, hacked hardware to see what it picks up, or gone out into the field to collect data where few people go.

If you're drawn to broadening beyond the purely technical, learning to work with prospects and clients, that's a plus too.

Location

Paris 11e, in a renovated late-19th-century industrial building down a quiet alley. Everyone has their own powerful GPU at their desk. You will not be hardware-poor.

Remote

Very limited for the first six months. This is a small team and a lot of what you'll learn happens in the room. After that it's negotiable, depending on how well you actually work remotely.

Language

We work in English; French helps but isn't required.

Offsites

We aim for two a year, most likely in Brittany near a surfing spot. They're optional but they're good fun.

Format

We don't have a single job to fill or a fixed description to hand you. Depending on who you are and what you've built, there might be room for an internship, a gap year or something more permanent.

Compensation

Depends on profile and experience, with the usual benefits (meal vouchers, health insurance, transportation pass, etc.).

Legal

Some roles include confidentiality and, where relevant, non-compete clauses, which we'll discuss openly at the offer stage.