Original job description
We are building autonomous spacecraft. This is the role that makes them think.
Ground control is a bottleneck. Latency, bandwidth, and mission complexity mean the future of spacecraft operations is onboard intelligence - systems that sense, reason, and act without waiting for a human uplink. We are building that future, and we need a Head of AI to own it end-to-end.
This is not a strategy role. You will set technical direction and get into the details - from sensor optics and compute hardware to perception models, inference runtimes, and the MLOps infrastructure that keeps everything improving. You will be the person who can look at a failure and know whether it's the model, the camera, the preprocessing pipeline, or the timing stack.
What you'll own
Autonomy stack architecture
You define how the full perception-to-action pipeline is structured: sensing → data acquisition → preprocessing → perception → state estimation → decision-making → control interfaces. You choose the algorithms, make the system-level trade-offs, and ensure the architecture survives contact with real spacecraft hardware.
Perception and computer vision
Object detection, segmentation, tracking, recognition, visual odometry, anomaly detection - you understand both classical CV and modern deep learning well enough to know which tool is right for which constraint. You can evaluate a model's failure modes, not just its benchmark numbers.
Edge and onboard AI
Spacecraft are resource-constrained environments. You will lead model compression, hardware-aware inference, ONNX/TensorRT deployment, GPU and accelerator integration, and real-time processing on embedded hardware. You know how to make a good model run fast on limited power and memory - and when to redesign the algorithm instead of squeezing the model.
AI infrastructure and MLOps
You build the systems that let the team iterate without losing reproducibility: dataset and annotation pipelines, automated training and evaluation, experiment tracking, model registries, CI/CD for AI, deployment pipelines, data and model versioning, monitoring, and active learning loops.
Research to production
You track emerging methods in CV, deep learning, RL, and autonomy. You design experiments to validate them under real spacecraft constraints - and you turn the ones that work into production-quality onboard software.
Team and technical leadership
You set technical direction, mentor AI/ML and computer vision engineers, establish development and review practices, and build expertise within the team rather than concentrating it in yourself. You own difficult failures, drive root-cause analysis, and influence decisions across engineering teams.
A day in the life
You might start the morning reviewing a perception failure on a hardware-in-the-loop test run - tracing it through sensor data, preprocessing logs, and model outputs to find the root cause. By afternoon you're in an architecture review for the next autonomy milestone, debating state estimation approaches with the GNC team. Later you're in a 1:1 with an ML engineer working through a model compression problem, and you end the day reviewing a pull request on the inference runtime. Slack is the primary async channel; the team runs two-week sprints with weekly technical syncs.
Must-haves
• Deep hands-on experience in at least one area of machine perception - computer vision, object detection/segmentation, tracking, representation learning, or sensor fusion
• Strong software and systems engineering background: Python, C/C++, Linux, GPU computing, containerised deployment
• Practical understanding of the physical layer of AI systems: cameras, optics, sensors, signal acquisition, compute architectures, embedded hardware
• Experience taking ML systems from research prototype to production: model optimisation, inference runtimes, edge deployment, monitoring, reproducible training, real-world debugging
• Demonstrated technical leadership: architecture decisions, mentoring, code and design reviews, communicating trade-offs, building team capability
Strongly preferred
• Reinforcement learning and learning-based control
• Robotics, simulation, and hardware-in-the-loop / software-in-the-loop testing
• Visual odometry / SLAM and state estimation
• Planning and control; robotics middleware
• Real-time systems and flight software
• Satellite or spacecraft systems; orbital mechanics and spacecraft dynamics
• Cloud platforms (AWS, GCP or equivalent) for ground-side AI pipelines and fleet-scale data processing
The engineering mindset we value
We practice first-principles engineering. Before adding a new model or more post-processing, we ask:
• Is this actually a model problem - or a sensor problem?
• Are we losing information in preprocessing?
• Is the bottleneck compute, memory, bandwidth, or latency?
• Can the algorithm be redesigned instead of patched?
• Do we have the right data to answer the question?
• What happens when the system leaves the laboratory?
If that framing resonates with how you work, we want to talk.
Backgrounds we consider
We are open to candidates from computer vision, robotics, autonomous vehicles, aerospace autonomy, embedded AI, industrial perception, defense and space systems, and research engineering. What matters most is the combination of deep technical expertise, broad systems understanding, and the ability to lead a technical team.
What success looks like in year one
• A coherent, documented technical architecture for the full autonomy stack
• A strong AI engineering team with clear development culture and practices
• Reproducible data and model pipelines in production
• Demonstrated AI capabilities on representative spacecraft hardware or high-fidelity simulation
• A clear, credible path from research prototypes to deployable onboard software
• The most important perception and autonomy risks identified and systematically being reduced
