Senior Machine Learning Engineer (Computer Vision for Earth Observation)

• We are looking for a Senior ML Engineer (f/m/x) to build and scale computer vision systems for Earth observation
• The core of the role is geometric computer vision on very high resolution satellite imagery: 3D reconstruction from stereo and multi-view data, and robust image matching and registration across sensors, viewpoints, and time
• These systems turn raw optical imagery into accurate 3D surface models and precisely aligned image stacks that downstream products can rely on
• Around that core, the role extends into broader CV problems such as segmentation, detection, and change analysis, and into making models run reliably in production, including under constrained compute where projects require it
• A central focus of your first project will be the generalization of stereo reconstruction methods: making learned stereo perform reliably across sensors, geographies, and acquisition conditions
• We have already done substantial modeling work here, and experience with synthetic data generation and sim2real transfer would be especially valuable in taking it further
• This is a balanced role: part applied research, part engineering, all impact
• The exact balance depends on your strengths, and we are open to profiles that lean more toward applied research or more toward engineering as long as the fundamentals are strong
• You’ll be part of Sektion 4, LiveEO’s government-solutions product team. Sektion 4 owns its roadmap and delivers funded R&D projects end-to-end, from research through to production, and sets its own technical direction
• You’ll collaborate with other LiveEO teams and with external research partners while retaining ownership of the team’s goals and deliverables
• You’ll also work closely with our data and annotation function to define labeling and quality guidelines and to close feedback loops on data quality across geographies and acquisition conditions
• As a Senior ML Engineer, you will drive the development of state-of-the-art computer vision systems that reconstruct 3D structure from, and robustly align, large volumes of satellite imagery
• Drive geometric CV development: design, train, and iterate on stereo/multi-view 3D reconstruction models and image matching/registration pipelines for VHR optical imagery (co-registration, alignment, robust correspondence), with clear ablations and measurable performance improvements
• Research to production: identify and adapt state-of-the-art approaches in 3D reconstruction, depth estimation, feature matching, and adjacent geometric CV (papers → prototypes → validated baselines), focusing on pragmatic wins under real constraints
• Tackle generalization head-on: close domain gaps in learned stereo across sensors, geographies, and acquisition conditions, including through synthetic data and sim2real transfer strategies
• Broader CV where projects need it: contribute to semantic tasks such as segmentation, detection, and change analysis that build on the aligned imagery and 3D reconstructions the core work produces
• Own EO data quality: standardization and preprocessing for high-resolution imagery (normalization/calibration, tiling, pairing and co-registration sanity checks, sampling/augmentation), plus dataset-quality diagnostics
• Build scalable pipelines: training and evaluation infrastructure across cloud and secure on-prem environments, with experiment tracking, reproducibility, and systematic failure analysis across geographies and acquisition conditions
• Deliver production-ready components: robust inference interfaces, model packaging, deterministic evaluation, and monitoring, plus, where relevant, adaptation of models to constrained or on-device compute
• Collaborate and communicate: work with the data annotation function on labeling guidelines and edge cases, with partner teams to turn model capabilities into validated deliverables, and with external researchers, presenting findings clearly and efficiently

Distributed computing with Ray; workflow orchestration with Prefect (or similar) is a plus

You enjoy working with complexity and turning ambiguity into structure

3D / photogrammetry tooling: NASA Ames Stereo Pipeline, MicMac, COLMAP; DSM generation is a plus

Hands-on experience with satellite / remote-sensing imagery is a plus

Comfortable working with researchers and presenting findings clearly and efficiently

You communicate clearly and collaborate smoothly within and across teams

Strong understanding of ML experimentation, versioning, and tracking

You take ownership and proactively push work forward

Strong Python engineering fundamentals with clean, maintainable code, and deep experience with PyTorch, implementing and training deep learning models at scale

Eligibility to obtain a German security clearance (Sicherheitsüberprüfung)

Broader geometric CV: structure-from-motion, SLAM / visual odometry, or neural 3D representations (e.g. NeRF, Gaussian splatting) is a plus. is a plus

Strong computer vision fundamentals (representation learning, supervision strategies, evaluation design) and practical debugging/optimization skills

Background in remote sensing, computer science, physics, or a related field, or equivalent practical experience. A PhD in one of these fields is a plus

Experience with PostgreSQL (or similar) is a plus

Experience with GDAL, Rasterio, GeoPandas, STAC is a plus

Experience deploying models under constrained compute or on edge devices (model compression, quantization, optimization) is a plus

Pragmatic mindset: you balance deep research with practical delivery

Experience with synthetic data generation and sim2real / domain adaptation for geometric vision tasks is a plus

Cloud platforms (AWS) and/or secure on‑prem / HPC experience (SLURM, Docker, DVC) is a plus

Experience with SAR alongside optical imagery, or familiarity with geospatial foundation models/ VLMs (self-supervised, contrastive, masked modeling) is a plus

Practical experience in at least one area of geometric computer vision: stereo/multi-view reconstruction, depth estimation, or image matching/registration