Research Scientist: Pre-Training

Enact is an early‑stage startup that is redefining the robotic technology stack to bring general‑purpose intelligence into physical systems. The company builds everything from model architectures and data pipelines to the actuators and sensors that feed those models, aiming to create a proprioceptive nerve system and a cortical layer that enables dexterous, force‑aware manipulation across thousands of tasks. In this Research Scientist: Pre‑Training role you will construct the foundational intelligence layer for robots. Your day‑to‑day work involves designing and running large‑scale pre‑training experiments for robot foundation models, specifying architectures, objectives and curricula that move AI‑driven manipulation beyond visual inputs to actual actions. You will own the full experimental pipeline, from data specification through training and real‑robot evaluation, while investigating scaling laws, data‑quality impacts and architectural trade‑offs. Building datamixes and evaluation sets for high‑precision tasks in both simulated and real environments, as well as researching methods to improve sample efficiency and robustness, are also core responsibilities. The position requires deep experience developing large‑scale predictive or generative models such as RSSM‑style, JEPA, transformer or diffusion/flow‑based architectures. You must have strong foundations in probability, statistics and machine‑learning theory, and be able to design rigorous experiments that separate genuine signal from noise. A PhD or equivalent research experience with a proven record in machine learning, computer vision, data‑science, robotics or related fields is mandatory, as is solid expertise in ML computing frameworks like PyTorch or JAX, including performance‑oriented optimization at scale. Additional assets include prior work with real robot hardware, leadership of multi‑node, multi‑GPU distributed training projects, experience designing or scaling data‑collection, annotation and mixing pipelines, and a history of open‑source contributions or released models and datasets. The role is based in central Zürich, offering a world‑class in‑person setup with top‑tier prototyping, robotics and technical infrastructure. Enact provides visa sponsorship and relocation support, competitive compensation, meaningful equity, and the chance to see your research deployed in real‑world robots from day one. What the role asks for: - Deep experience with large‑scale predictive or generative models - Strong probability, statistics and machine‑learning fundamentals - PhD or equivalent research experience with proven record - Proficiency in PyTorch or JAX for large‑scale training - Prior work with real robot hardware (nice‑to‑have) - Led multi‑node, multi‑GPU distributed training efforts (nice‑to‑have) - Designed or scaled data collection and mixing pipelines (nice‑to‑have) - Open‑source contributions or released models/datasets (nice‑to‑have)