<h2>Company</h2>NestAI is a European AI lab for defence delivering the adaptive operating system for modern battlefield operations. We develop adaptive intelligence for unmanned and command systems: AI that continuously learns from operational data and adapts to changing conditions.NestOS is our open, modular and interoperable platform that enables this at scale and keeps capability evolution under sovereign European control. Founded in 2025 by Peter Sarlin, whose previous company Silo AI became Europe's largest AI acquisition, NestAI brings together over 200 engineers and scientists and partners with leading European defence forces and industry.<h2>Role Description</h2>You'll join the synthetic data team inside our ML Platform chapter, the group responsible for generating training data that doesn't exist anywhere else yet. Real-world imagery for object detection in defence contexts is scarce by nature: the situations, assets, and conditions our models need to recognize are rare, hard to capture safely, or simply haven't happened yet. This team builds the generative pipelines that close that gap, generating visual data under controlled conditions for scenarios that are difficult to capture directly. This role is for someone who has trained generative models themselves, not just fine-tuned or called pretrained ones, and who cares as much about whether synthetic data holds up in real-world model performance as about whether it looks convincing.<h2>Day-to-Day Responsibilities:</h2>Since high-quality real-world imagery for our use cases is scarce, this team's job is to generate data that closes the gap without compromising downstream detection performance.<ul><li>Build and improve generative pipelines that produce synthetic image and video data for object detection models, covering scenarios where real data is scarce or unavailable.</li><li>Train and fine-tune diffusion-based generative models, including LoRA fine-tuning, to produce increasingly realistic and varied synthetic data.</li><li>Combine generative approaches with other data sources to boost realism and variation and increase variation in the data produced.</li><li>Validate synthetic data against real-world outcomes, proving it improves model training, not just that it looks good.</li><li>Track the fast-moving generative model landscape and evaluate new open-source models as they emerge.</li><li>Work closely with the wider ML Platform chapter and simulation team on shared data infrastructure.</li></ul><h2>Qualifications:</h2><ul><li>Several years of hands-on experience training generative models yourself, not only fine-tuning or calling pretrained APIs.</li><li>Experience creating synthetic data in domains where real data is limited — image-based fields like medical, hyperspectral, or multispectral imaging are a strong analogue, even outside defence.</li><li>Deep familiarity with diffusion models and Python-based ML training frameworks, including LoRA training and the Hugging Face Diffusers library.</li><li>Judgment for when synthetic data is realistic enough to actually matter for training, and when it isn't, however good it looks.</li><li>Comfortable working with imperfect information in a research area that shifts month to month.</li></ul>Nice to have:<ul><li>Experience with Nvidia Omniverse or other simulation environments, and combining simulation output with generative models.</li><li>Exposure to generative world models.</li><li>Background in a field where generating photorealistic, cost-efficient synthetic scenes at scale mattered (e.g. automoti

Synthetic Data Generation Engineer

About this role

EU Requirements

Job Details

Contact

Similar Jobs