Comand AI's mission is to build next-generation C2 software with real users in real deployments. We are fully indexed on the success of every single user, which means our research has to work in the field, not on a benchmark.
As a Computer Vision Engineer, you own the hardest perception problems in the stack. The job is to crack the scientific problems that prevent us from turning multi-source sensor data (satellite, drone, ISR, OSINT) into real-time operational clarity. Your models ship into production on short cycles, used by people for whom bad output has real consequences.
In practice:
Design and train architectures for object detection and segmentation on satellite and aerial imagery, and classification across complex image types
Drive our applied research agenda: track the state of the art rigorously, form hypotheses, iterate fast
Push models to production readiness: optimize for inference speed and robustness under distribution shift, which will happen in field conditions
Work closely with data engineers and software engineers to move work from research to production without losing what matters
Spend time with end users and domain experts to understand what "good" actually means in their context
We want scientific depth combined with the instinct to ship. Not a publication count, but evidence of both: a GitHub and Google Scholar that tell a coherent story. A recent PhD is one path in. A more experienced researcher with a strong collaborative instinct and a track record of shipping is equally welcome.
Must-have:
PhD in AI, deep learning, or computer vision, or 5-10 years of serious applied research with a track record to back it up
Strong deep learning expertise, non-negotiable. Computer vision experience strongly preferred
Solid Python, Git, and PyTorch. JAX/Flax, scikit-learn, and Numpy are real pluses
Hands-on experience with Transformer and multimodal architectures applied to visual data
EU citizenship (required by our defense contracts)
English fluency
A genuine appetite for cross-functional work: you explain your technical choices, co-design with product and operational teams, and take ownership of outcomes rather than just models
Nice to have:
Applied work with satellite, aerial, or drone imagery
Experience deploying models to constrained or sovereign environments
SQL/PostgreSQL
Prior work in defense, dual-use, or operational tech
Geospatial data processing background
Talent screening: 30 min, online. We talk through your background and motivation.
Research deep dive: 60 min, online. We evaluate the depth of your expertise and how you reason through hard CV and deep learning problems.
Technical interview: 60 min, online. Coding and applied ML problem-solving. A second round may follow if we want more signal.
Onsite in Paris: half day. Analytical decomposition then a short project presentation.
Roundtable: hiring committee
