Robotics / ML Engineer

Robotics / ML Engineer In short: funded, revenue-generating deep-tech AI company, still in stealth. Growing the engineering team. This is a hands-on build role for someone early in their career. You will spend real time with physical hardware. • Location: Boston or Cambridge. Regular in-person time required for hardware work. • Comp: $150,000 to $200,000 base, plus equity. Health, dental, vision, 401(k) with match. • Experience: 1 to 4 years, or a recent PhD. Research output (papers, preprints, maintained open source) is a strong plus. What the company does • Spun out of a top research lab, backed by a specialist deep-tech investor. Paying customers and real revenue while still in stealth. • Most AI today means bigger and bigger models trained offline. This team is doing the opposite: small systems that learn as they go, from live data, on a fraction of the usual compute. It works in the real, messy physical world. Customers are robotics companies, AI labs, and pharma. What you'd do • Take learning systems from research code to running on real robot hardware, and debug them when they fail on the robot rather than in the notebook • Build evaluation and testing infrastructure: benchmarks, regression tests, reproducible experiment tracking • Run experiments, analyse results properly, and turn them into decisions the team can act on • Support customer deployments and integrations on their hardware • Profile and optimise models to run inside real compute and latency budgets Requirements: You should be able to point to specific work for most of these: • Physical hardware deployment. You have personally run learning or control systems on a real robot in a real environment: a manipulator, a mobile base, a drone, a vehicle, a legged platform. Coursework and simulation-only projects do not count. Be ready to tell us what broke and how you fixed it. • Python at a production standard. Not just scripts. Tests, packaging, code someone else can run. • C or C++ for anything touching the control loop or embedded side. • PyTorch or JAX at depth. You can write custom training loops, debug a model that trains but does not work, and read someone else's research code. • Real depth in at least one of: reinforcement learning, control theory, robotic manipulation, or computer vision for robotics. Depth means you have implemented the methods, not just used a library. • Experimental rigour. You control variables, you notice when a result is too good, you can tell a real improvement from noise. Strongly preferred: These are the areas the work actually sits in. Experience in any one of them will move you up the list significantly: • Online, continual, or adaptive learning. Systems that keep updating after deployment, not a frozen model trained once. Includes lifelong learning, test-time adaptation, and anything dealing with distribution shift on live data. • Compute efficiency. Getting models to run inside tight budgets: quantisation, pruning, distillation, onboard or edge inference, real-time constraints on limited hardware. • Robotic manipulation. Grasping, dexterous control, contact-rich tasks, learned policies on real arms. • Risk-aware or safe control. Uncertainty quantification, robust or safe RL, control under model error, anything where failure has a physical cost. • Signal processing or applied mathematics. Sparse methods, sampling theory, information theory, or similar. Also useful • ROS or ROS2 • Real-time and embedded systems, low-level control • Sim to real transfer • Publications at robotics or ML venues, or open source people actually use Thanks, Team Amberes