Founding Machine Learning Engineer

ABOUT THE ROLE As the founding machine learning engineer at an early-stage AI company, you will build the ML function from the ground up. You will shape post-training and agent systems, connecting research with reliable production infrastructure. WHAT YOU'LL DO - Structure, filter, and score experimental trajectories for training data pipelines. - Design and implement evaluations and benchmarks for model reasoning, planning, and experimental progress. - Build reliable agent environments, tool interfaces, observability systems, and replay infrastructure. - Establish validation and provenance tracking to support trajectory and data quality. - Set priorities for the ML roadmap, experiments, systems, and hiring. - Guide the technical direction of the ML team as it grows. WHAT WE'RE LOOKING FOR - At least 3 years of professional machine learning engineering experience building or delivering production ML systems. - Professional experience in post-training or model evaluations, including hands-on work structuring, filtering, and scoring training data. - Production software engineering experience with Python and systems-level programming for ML infrastructure. - Experience designing and implementing ML evaluation frameworks and benchmarks. - Strong knowledge of trajectory data, reward structures, tool use, and agent decision-making. - Experience building agent environments, replay and debugging tools, RL training systems, or data validation and observability infrastructure. - Reinforcement learning implementation and tuning experience is valuable, as is experience in frontier AI research or scaled post-training and evaluation teams. - Comfort owning priorities and working across research and production in an ambiguous environment. COMPENSATION & BENEFITS Base salary: $100,000 to $250,000 USD annually. Visa sponsorship is not available. LOCATION On-site in Munich, Germany. Other listed locations include Zurich, Switzerland, and San Francisco, California; remote arrangements may be discussed.