Dusejovsky Kamil, e-mail: [email protected]
We are looking for an experienced professional who will lead, scale, and advance our training pipeline. Better models, larger datasets, more robust production stack, every percentage point of quality in strength, intent, and policy.
What you will do
Lead the training pipeline at scale. Deploy improved skill models to real robots on a weekly cadence.
Advance model quality. Architecture, loss function design, tokenization, data curation, augmentation, evaluation. Own ablations.
Own training infrastructure. Multi-GPU, distributed runs, checkpointing, experiment tracking, automated evaluation.
Approach data as a first-class concern. Curation, labeling, balancing, synthetic augmentation, sim-to-real.
Be in the lab. Sit with the hardware and capture team. Close the loop between model and robot.
Who we are looking for
Experience in computer science, electrical engineering, robotics, ML, or a related field. Or completed work that demonstrates the same depth.
Native work in PyTorch. Comfort with CUDA, multi-GPU, distributed training, mixed precision.
Hands-on experience with at least one of: imitation learning, behavior cloning, diffusion policies, vision-language-action models, RL on robots, sim-to-real.
Fluency across the modern robotics learning stack: large multimodal foundation models, photorealistic simulators, ROS-class middleware, modern policy architectures.
Strong foundations in signal processing, sensor fusion, and high-dimensional time series modeling. Multi-rate synchronized sensor data doesn't intimidate you.
You take data seriously. Quality, balance, and curation are your job, not someone else's.