Representative role.This is a composite of searches we run regularly at this level and stack. If your background matches, apply anyway — we’ll match you to similar live briefs.
A scaling defence-tech company is hiring engineers who can actually ship firmware *and* models. The role sits between ML R&D and embedded systems — you'll take models from Jupyter to Jetson / STM32 and make them run in real time on real hardware.
This is not a SaaS ML role. You'll be quantising, pruning, profiling on-target, and working closely with sensor-fusion and autonomy teams to hit latency budgets measured in milliseconds, not seconds.
Australian citizenship required — the role is clearance-eligible from day one, with NV1 sponsorship after probation.
What you’ll do
- Convert PyTorch / ONNX models to deployment formats (TFLite Micro, TensorRT, CoreML) and optimise for target hardware
- Own the on-device inference pipeline end-to-end: quantisation, pruning, benchmarking, memory profiling
- Work with firmware engineers on sensor integration (IMU, LiDAR, camera) and real-time scheduling
- Write C/C++ where Python cannot reach — model runtime glue, custom operators, hardware drivers
- Establish the team's embedded-ML patterns: CI for on-target tests, model versioning, OTA update flow
What you bring
- 5+ years shipping production ML, with at least 2 years deploying to resource-constrained hardware
- Fluent in Python + C/C++. Comfortable reading disassembly and profiling at the instruction level
- Deep familiarity with at least one embedded ML stack: TFLite Micro, TensorRT, ONNX Runtime, STM32Cube.AI
- Experience with quantisation (INT8/INT4), pruning, or distillation — not just ran a notebook once
- Australian citizen (required for clearance eligibility)
Nice-to-haves
- Prior defence, aerospace, or robotics industry experience
- ROS 2 / real-time middleware familiarity
- Existing Baseline or NV1 clearance
- Sensor fusion or SLAM background