Thesis

As a Member of Technical Staff, you will design and build the architectures, evaluation loops, training systems, and orchestration layers that allow machine learning research to compound. Your work will become the foundation for neo labs applying AI to materials discovery, robotics, drug discovery, climate science, and other frontier domains. What You’ll Work On Autonomous R&D systems Design workflows for hypothesis generation, experiment planning, model training, evaluation, debugging, and iteration. The hill-climbing engine Build systems that search large spaces of architectures, hyperparameters, datasets, losses, and training procedures, using each result to improve the next experiment. Frontier AI infrastructure Engineer the APIs, schedulers, queues, storage, and observability that run many experiments reliably in parallel across models, datasets, and GPUs. Recursive improvement loops Create systems where better models produce better experiments, and better experiments produce better models. Who You Are You have trained real ML models You have hands-on experience training machine learning models, whether in deep learning, reinforcement learning, evolutionary search, optimization, or related areas. You are strong at systems You have built reliable backend, cloud, distributed, or infrastructure systems. You can reason about scalability, fault tolerance, orchestration, observability, and performance. You have research taste You have research experience in CS, ML, AI, or a related field. Publications at top conferences like ICLR, NeurIPS, or ICML are a plus, but we care more about your ability to reason from first principles, run good experiments, and make progress on hard problems. You move between research and engineering You can design experiments, write training code, debug infrastructure, and ship systems that actually work. You are driven to explore the frontier You want to accelerate scientific discovery and are comfortable exploring uncharted directions with minimal supervision. The Stack We use whatever safely and rapidly scales the system. Today that includes Python, Rust, PyTorch, TypeScript, GPU orchestration, cloud/backend infrastructure, evaluation harnesses, experiment tracking, and the systems required to turn research into a compounding loop.