Machine Learning Engineer

Mariana Minerals is building machine learning systems that control mineral refining facilities, with a strong focus on reinforcing learning in physically realistic simulators and closing the gap to real plant data. This is an onsite role in Ann Arbor, MI, where your work moves from experimentation to production. The salary range is $120,000 - $180,000 per year.

What you’ll do

• Run reinforcement learning experiments in physically realistic simulators of mineral processing operations, then translate results into improved controllers.
• Build and refine training environment components, including reward functions, observations, and action logic, with guidance from senior engineers.
• Train control models, track and interpret performance, and investigate why a model underperforms.
• Close the loop between simulation and reality by comparing model behavior to real plant data and flagging where simulated physics diverge.
• Write clean, well-tested code and contribute to the services that deploy models into production.
• Partner with process and chemistry experts to understand the unit operations being modeled.

What you bring

• 2-8+ years of experience (including internships or research) in machine learning, reinforcement learning, or scientific computing, or a strong recent graduate with demonstrated project depth.
• Solid grounding in machine learning fundamentals, with working knowledge of modern deep learning; reinforcement learning experience is a strong plus.
• Proficiency in Python and comfort reading and debugging an existing codebase.
• Curiosity about physical, industrial systems and willingness to learn chemistry and process engineering from experts who challenge assumptions.
• A self-starter who asks good questions, ships, and escalates blockers early.

How the technology is used

The stack uses the same reinforcement learning toolkits that power self-driving vehicles and humanoid robots, adapted for autonomous, short-interval control of mineral refining circuits. Models adjust operating set points and configurations in real time, optimizing across lithium recovery, reagent consumption, energy intensity, and equipment uptime simultaneously. Training happens in physically realistic simulators of process units, then performance is validated against real plant data before anything is introduced to live equipment.

Why this matters (and why you’ll like it)

• We own the projects, generate the data, and close the loop.
• Each facility helps make the software smarter, and the next one faster and cheaper.
• Mining is one of the last major industrial sectors that hasn’t been rebuilt with modern software.
• Your work directly shapes how critical minerals are produced at scale in the coming decades.

Culture at Mariana Minerals

• Extreme Ownership – take full responsibility for outcomes and drive toward solutions.
• Engineer Out Requirements, then Automate – simplify, optimize, and automate for scale.
• Share Your Legos – collaborate openly, share knowledge, and empower each other to build bigger, better solutions.