Imagine this. You are working on a precision inertial sensor where even after control and conventional compensation, a small residual error remains. We want to find out how much of that error can genuinely be predicted and removed using machine learning.

As a Machine Learning Engineer, you will own that investigation. You will build learned compensation models, benchmark them against a strong classical baseline, and determine where ML delivers measurable value and where it does not. This is a hypothesis to test rigorously, not a predetermined solution.

At Destinus, we are revolutionizing the defense industry with cutting-edge Unmanned Aerial Vehicles (UAVs). Our innovative technologies are designed to meet the unique demands of modern defense operations, delivering unparalleled speed, precision, and cost effectiveness. Destinus partners with government agencies and defense organizations worldwide to provide advanced solutions for mission-critical operations, enabling a new era of efficiency and technological superiority. Join us in shaping the future of defense with groundbreaking aerospace innovations.

What You'll Do

• Build ML models that predict residual sensor error using observable signals including temperature, thermal gradients, quadrature amplitude, drive signals, and sensor diagnostics
• Define rigorous validation protocols across unseen thermal profiles and physical sensor units to demonstrate genuine generalisation
• Benchmark learned approaches against a tuned classical baseline combining per-unit thermal compensation and adaptive Kalman filtering
• Quantify the observability boundary and identify which errors are predictable from available measurements and which are fundamentally outside the model's reach
• Train and evaluate models offline using sensor characterisation data, separating meaningful physical correlations from artefacts and overfitting
• Work with FPGA and DSP engineers to translate successful approaches into lightweight, frozen models suitable for low-latency embedded deployment
• Communicate results clearly, including when the evidence shows that a classical approach remains the better solution

Requirements

• B.Sc., M.Sc. or PhD in Computer Science, Applied Mathematics, Applied Physics, Machine Learning, or a related technical field
• Strong applied machine learning experience with time-series, regression, sensor, or instrumentation data
• Strong Python or MATLAB skills and experience with PyTorch, scikit-learn, or equivalent modelling frameworks
• Experience working effectively with modest, high-value datasets where validation strategy and data quality matter as much as model architecture
• Strong understanding of model validation, generalisation, overfitting, feature engineering, and experimental design
• Enough physics and signal-processing knowledge to challenge whether a discovered relationship is physically meaningful or simply an artefact in the data
• C/C++ skills are highly desirable
• Experience with sensor calibration, metrology, inertial sensing, or similar physical measurement systems is a strong advantage
• Experience deploying ML models to embedded or resource-constrained targets is a plus
• Exposure to aerospace, defence, robotics, or other high-performance engineering environments is a plus

Who You Are

You care more about whether a model works than whether it is fashionable. You are comfortable challenging your own results, designing experiments that are difficult to fool, and saying when the data does not support the hypothesis. You combine strong ML skills with enough curiosity about physics and signal processing to understand what is happening behind the dataset, and you enjoy turning experimental results into clear engineering decisions.