Machine Learning Engineer for Edge / Embedded Intelligence Platform Marine Platform

We're an early-stage startup building an edge-first intelligence platform that runs entirely on-device, no cloud. Here's the part that matters: we deliberately use low-cost, commercial and consumer-grade instruments and then derive reliable, high-quality intelligence out of them through clever software modeling and correction, instead of buying expensive purpose-built sensors. That approach is core to what makes us different, and it only works if the person building it deeply understands the hardware we're running on and knows how to program against its real-world limits. So this role is equal parts two things. First, you understand the hardware in front of you: the edge compute, the instruments, the vessel networks, their quirks, timing, resource limits, and failure modes, and you write software that accounts for all of it. Second, you think out of the box. The problems here rarely have a textbook or off-the-shelf answer. We need someone who reasons from first principles and invents the approach, not someone who plugs in the standard library and hopes. To be clear, this is not a hardware-bring-up or board-design role. You won't be building hardware. But you have to understand it cold and know how to squeeze reliable results out of it in software. What you'll do: Understand our hardware stack at a deep level (edge compute, instruments, vessel networks) and write code that programs directly against its constraints, timing, and failure modes Write acquisition and decode code that pulls raw data off vessel networks (CAN bus, Modbus, NMEA 2000) at the wire level Take imperfect, low-cost, consumer-grade instrument data and turn it into reliable measurements through modeling and correction in software (this is where most of the inventive work lives) Build the ML and analytics that convert that data into anomaly detection, root-cause and downstream-effect analysis, and predictive maintenance Optimize and deploy models to run on-device on constrained edge compute, fully offline Fuse many independent onboard systems into one coherent picture Who we're looking for: A genuine out-of-the-box thinker who solves problems that don't have a known answer, works around constraints, and does more with less Deeply hardware-literate: you understand the hardware you run on and know how to program for it, not just consume it through a framework Resourceful and scrappy, comfortable in a fast-moving early-stage environment with real ambiguity and limited hand-holding Must-haves: Strong Python and solid ML engineering fundamentals Real, hands-on experience acquiring and decoding data off CAN bus at the wire level (writing the acquisition code yourself, not just reading already-decoded messages) Deep comfort programming for and around real hardware constraints (edge/embedded compute, timing, resource limits) Real-time / streaming time-series and sensor-data processing Anomaly / fault detection and predictive maintenance A track record of creative, non-obvious solutions to genuinely hard problems Nice to have: Marine or NMEA 2000 experience Getting reliable results out of imperfect or low-cost sensors Sensor fusion and digital-twin work Safety-critical or deterministic-systems exposure (we run a hard real-time safety core alongside the intelligence layer) Familiarity with robotics middleware such as ROS2, though this sits above the bus, not in robotics