Machine Learning Engineer, Automated Calibration Services

We are hiring a Machine Learning Engineer into Automated Calibration Services, the team that keeps qubits inside spec while programs are running. Physicists design the calibration protocols; you will build the models that decide when they run, drive them from a schedule, a compiler request or a monitoring event, and establish afterwards whether they worked.

The open questions are inference problems: which qubit is about to drift out of spec, which routine recovers it fastest, whether an optimiser can replace a parameter sweep in a fraction of the device time, and whether a change in a measurement is a device defect or noise. Device time is scarce, so an unnecessary calibration costs program execution.

Model output triggers a physical operation on a device. Each decision needs an uncertainty estimate, a fallback when confidence is low, and instrumentation to establish afterwards whether the qubit improved. Data is expensive, the process is non-stationary, and a model trained on last month's device may not hold.

Based at our Sydney facility, you will work daily with the physicists who own the domain knowledge and the engineers who run the calibration stack. This is a role for someone who wants their models to drive real hardware, and who treats uncertainty quantification and validation as the substance of the work rather than an afterthought.

• →Build models that predict qubit drift and device health, and turn those predictions into decisions about which routine runs and when
• →Replace exhaustive parameter sweeps with sequential optimisation, using Bayesian optimisation, active learning or comparable methods, to reach the same tuning outcome in less device time
• →Apply computer vision and signal analysis to device measurement data where those methods outperform simpler alternatives
• →Build the feature and telemetry pipelines out of the calibration store that these models depend on
• →Quantify uncertainty, so the orchestration can separate a high-confidence recommendation from a low-confidence one and act accordingly
• →Validate models against held-out device data, including the case where the device has changed since training
• →Deploy models into the calibration loop with monitoring, fallback paths and a kill switch
• →Distinguish real drift and defects from measurement noise, and set the thresholds that trigger action
• →Work with physicists to encode what they already know as priors and constraints rather than making the model learn it twice
• →Instrument the loop so the effect of a model-driven calibration on qubit performance is measurable after the fact
• →Feed device characterisation and noise models back to Compiler Services and Control & Error Correction
• →Document models, assumptions and failure modes, so an automated decision can be explained to the owner of the affected device

• 4+ years applying machine learning in production, ideally where the output drives a physical system
• Strong Python and the scientific stack: NumPy, SciPy, pandas, scikit-learn, and PyTorch or JAX
• Sequential decision-making under expensive experiments: Bayesian optimisation, Gaussian processes, active learning or bandits
• Time series modelling, and anomaly or drift detection on real telemetry
• Uncertainty quantification, and knowing what a calibrated confidence interval is worth
• Validating models where data is scarce, correlated and non-stationary, and recognising an implausibly strong result
• Deploying models into a loop with monitoring, alerting and a fallback path
• Working with scientists on a problem where the domain knowledge is theirs and the automation is yours
• Numerical work: curve fitting, parameter estimation and optimisation
• Git workflow, code review and testing applied to research code
• Clear technical writing

• Computer vision on instrument, microscopy or measurement data, with OpenCV or PyTorch
• Reinforcement learning, or control policies learned rather than specified
• Qubit calibration, tune-up or device characterisation, on any qubit modality
• Control theory and closed-loop feedback
• Laboratory instrument control: QCoDeS, Labber, pyvisa or SCPI
• MLflow or comparable experiment tracking, and reproducibility practice
• Inference under a latency budget, including on edge hardware or FPGA
• Workflow orchestration such as Airflow, Dagster, Prefect or Temporal
• Quantum computing exposure of any kind (no physics degree required)
• Publications, open source contributions, or an experimental automation project you are willing to talk through

SQC is an equal opportunity employer. We value diverse perspectives and experiences, and encourage applications from candidates who may not meet every listed requirement. If you’re excited about the role and believe you can contribute, we encourage you to apply.

This position may require access to export-controlled information or technology. Employment may be subject to applicable export control laws and may require eligibility assessment based on factors such as nationality, citizenship, or residency, and, where necessary, obtaining relevant export licenses or approvals.

SQC was founded by renowned physicist and materials scientist Michelle Simmons, who pioneered the field of atomic electronics, including the development of the world’s first single-atom transistor and the first integrated circuit built with atomic precision. Our Chair, Simon Segars, former CEO of Arm, is a leader in the semiconductor industry and was instrumental in developing the processors that powered the mobile computing revolution.