Senior Product Manager, Machine Learning

About Machina: Engineering moves at software speed. Manufacturing doesn't. Yet. Machina Labs is changing that. We build intelligent, software-defined factories that produce complex metal structures directly from digital design. By integrating advanced metal forming, robotics, and automated production inside a flexible factory architecture, we enable customers to move from prototype to production in weeks, not years. Backed by Lockheed Martin, Toyota, and NVIDIA, we're building the manufacturing infrastructure that defense, aerospace, and advanced mobility programs will run on. If you want to work on hard problems that matter and see them fly, drive, and defend, this is the place.

About the Role The hardest part of Roboforming is generating a path that will produce the highly accurate part you set out to make. The metal does not want to be formed. It springs back the instant the tool moves past it, and once formed it is packed full of residual stress that wants to pull the part out of spec at any change in environmental conditions. The path we generate is never the part we get, so the real task is finding the path that delivers the part we are after. Today we find that path by iterating. We form a part, measure the error, and adjust the path to compensate. This works, but it is expensive. Every run cost material, labor, and robot time. To make our manufacturing accessible to broader consumer markets we need to reach the objective part tolerance in fewer trials. We get there by predicting springback before we form. Traditional solvers cannot get us there. Off-the-shelf FEAs are built for a handful of known loads, not millions of separate small bends. Instead, we need to learn the behavior. This role owns the products that let us do that: machine learning across all the parts we have formed, and GPU based physics simulations where we simulate complex physics. Planned Areas of Focus Springback Prediction You will own the models that predicts how the part will spring back and pre-compensates the path, so the very first formed part lands close to the target shape. Input Parameter Selection You will own the model that tells users which input parameters to pick during path planning, so the part forms optimally. This process is currently dependent on experience and tribal knowledge in a way that doesn’t scale. Data and Simulation Environments You will own the data sets necessary to train all ML models. This includes exploring historic data as well as working closely with the R&D team to plan experiments and runs necessary to fill in gaps in the data. Responsibilities Own the machine learning and simulation roadmap. Define priorities, make tradeoffs, and communicate direction to engineering and leadership.