Full-time · On-site (Paris Marseille Amsterdam)
The sea is the physical layer of the economy—the conduit for our energy, communications, and trade. It is also Europe’s exposed flank. Pelagon is building the answer.
Founded by Aymeric (serial deeptech entrepreneur; founded & exited Scortex, Leading AI solution for manufacturing quality) and Gauthier (Navy officer and McKinsey), Pelagon is building a foundational multimodal acoustic model for autonomous maritime protection (underwater and surface) in service of European sovereignty. It will be deployed on a distributed, autonomous network of vectors (UUVs, buoys, sailbuoys, seabed…) to provide real-time understanding of underwater threats.
France and Europe have world-class acoustic engineers and a long history in underwater warfare. By bridging this expertise with modern physical-AI strategies (as seen in autonomous driving and robotics). Europe is still missing it’s naval NeoPrime, Pelagon aims to become to lead the way in autonomous marine protection—from detection to interception—partnering with leading navies, primes, and neo-primes, while also operating its own fleet as a service.
Pelagon is backed by Hexa, the pioneer startup studio in Europe, which has launched 35+ companies, including unicorns like Front, Aircall, and Spendesk, combining €230m+ ARR and €750m raised.
Why this matters
The sea is the physical layer of the economy — and Europe's most exposed flank. Submarines, drones and threats move through it unseen because the ocean is opaque by default. Pelagon exists to change that: to give Europe a real-time, sovereign understanding of what moves beneath the surface, from detection to interception.
At the core of that mission is a maritime foundational perception model — Pelagon's equivalent of the perception stack that made autonomous driving possible, but for the underwater domain. As an AI Engineer, you own the models at the heart of it, from research to deployment on a distributed, autonomous fleet of UUVs, buoys, sailbuoys and seabed sensors.
What you'll do
Design, train and ship state-of-the-art multimodal and acoustic models for underwater detection, classification and localization.
Build the training pipeline end to end: self-supervised pre-training on large unlabeled acoustic corpora, then fine-tuning for specific threats.
Fuse acoustic, kinematic and other sensor modalities into a single representation that holds up in low-SNR, real-world conditions.
Shape Pelagon's data strategy — how we capture, curate and compound proprietary data from our own fleet into a durable advantage.
Distill and optimize models to run on the edge: compute- and power-constrained hardware aboard autonomous vectors at sea.
Close the loop with sea trials — turn field data and failure cases into the next iteration.
What we look for
Strong track record training models deployed in production, ideally on audio, signal, time-series or other non-image modalities.
A physical-AI mindset: you've worked where models meet messy sensor data and real hardware, not just clean benchmarks.
Instinct for data strategy — you treat the data flywheel as a core part of the model, not an afterthought.
Confortable across the ML stack: data, architecture, training, evaluation and deployment. Bias for shipping; you measure success by what works at sea.
Bonus: background in signal processing, or experience in defense, maritime, robotics or autonomous driving.
Why join now
Founding ownership. You'll be one of the first engineers and will define the technical foundation of the company. Meaningful equity comes with that.
A mission with weight. This is hardware that protects critical infrastructure and Europe's maritime space.
Real things, in the real ocean, fast. You'll see your work in the water within weeks, not years.
AI-first, at the edge, in the real world. We train our own models and run them where it counts: in the water, on the device. You'll be solving the edge-ML and power problems most companies only talk about.
A sovereign, European deeptech at the inflection point of a market with strong NATO and EU policy momentum behind it.
Skills Required
- Proven track record training models deployed in production
- Experience with audio, signal, time-series or other non-image modalities
- Experience designing and implementing self-supervised pre-training and fine-tuning pipelines
- Expertise in sensor fusion across acoustic, kinematic and other modalities
- Experience distilling and optimizing models for compute- and power-constrained edge hardware
- Physical-AI mindset: experience working with messy sensor data and real hardware
- Strong data strategy instincts for capturing, curating and compounding proprietary fleet data
- Comfortable across the ML stack: data, architecture, training, evaluation and deployment
- Background in signal processing
- Experience in defense, maritime, robotics or autonomous driving
- Experience closing the loop with field trials and iterating from failure cases
What We Do
Hexa is a startup studio. It all started in 2011 with startup studio eFounders, which pioneered a new way of entrepreneurship, became a reference in the B2B SaaS world, and launched over 50 companies including 3 unicorns (Front, Aircall, Spendesk). In 2022, eFounders became Hexa, to scale our ambitions further and launch startups in new verticals: AI, web3, fintech, future of work, health. With Hexa, we’re on a mission to democratize a new way for entrepreneurship, one centered around building companies in teams. In numbers? We’re an entrepreneurial ecosystem made up of: - +40 companies launched - +2800 jobs created - 3 unicorns (Spendesk, Aircall, Front) - +$4.5 billion combined companies valuation
