About the role

Our control systems can only act on what the plant knows about itself, and this role is responsible for most of that knowledge. The work includes:

• Reinforcement learning agents that learn to control our processes in simulation

• Time-series models for forecasting and detection

• Vision systems that measure things no instrument can

• Soft sensors that estimate lab results we can't get back fast enough

• Classifiers that flag faults early

You'll set technical direction for applied ML across our sites and build much of it yourself, from data to deployment. The scope is wide on purpose. We care more about range and good judgment than deep expertise in any single technique.

What you'll work on

• Deciding where ML is worth it. Some problems need a model and many don't. You'll help us put effort where it improves recovery, uptime, or cost.

• Time-series work: forecasting, anomaly and excursion detection, predictive maintenance, and soft sensors.

• Computer vision on circuits, conveyors, and equipment. Expect poor lighting, vibration, and dirty lenses.

• Classification and diagnostics for faults, product quality, and process state. Labels are rare, classes are imbalanced, and some mistakes cost far more than others.

• Setting the deployment bar for each model and running the validation that gets it from a notebook into the control room, where operators rely on it during their shift.

• Writing the team's modeling standards, reviewing designs, and mentoring engineers at every level.

• Working closely with two teams:

  • The autonomy and controls team, whose controllers use your predictions.

  • Data engineering, who own the platform underneath.

• Reinforcement learning in our process simulators, training agents to learn how to control each circuit and taking what works from simulation onto the real plant.

What we're looking for

• 8 to 10+ years in ML engineering, or 6+ with clear org-level technical leadership. You've put models into production in the physical world and kept them running.

• Shipped work in at least two of these areas: time-series, computer vision, reinforcement learning, and classical supervised learning. You can pick up a third quickly, and you can tell which one a problem actually needs.

• Experience with industrial sensor data: historians, SCADA/DCS, OPC-UA, or similar.

• Experience with scarce labels, drift, and rare events. You report how uncertain a prediction is, not just a single number.

• You work well with roboticists, chemists, metallurgists, process engineers, and geologists, and you can bring technical leadership along with you.

• You like building. Staff engineers here ship.

Why join

We own our projects and generate our own data, so what you build feeds straight back into operations. Each new facility adds to that data and makes the next one quicker and cheaper to build.

In most applied ML jobs, someone hands you a cleaned dataset and a metric. Here you get a plant. You figure out what needs measuring, build the thing that measures it, and see whether the numbers move.

Our culture is built on four principles: