AI Infrastructure Software Engineer — CosmosLab
Core
Design, assemble, and improve infrastructure for large-scale AI training (pre-training, SFT, RL post-training) supporting Physical AI world foundation models.
Role type
Senior IC AI Infrastructure Software Engineer
Builds
Training infrastructure, control planes, distributed training backends, inference/evaluation stacks, and RL simulation environments.
Domain
AI Infrastructure / Distributed Systems / Robotics Simulation
Deliverable
production ML models
Required skills
Large-scale distributed systems development, AI training/inference infrastructure, Python, debugging/triage across stack, software engineering practices (testing, CI), system-level optimization
Preferred skills
RL/post-training infrastructure (PPO/GRPO/DPO), simulation/robotics integration, DL framework internals (PyTorch FSDP/DTensor, Megatron), C/C++/CUDA
Technologies
PyTorch, Megatron, CUDA, C/C++, Python
Responsibilities
Create and implement training infrastructure spanning pre-training, SFT, and RL post-training; Develop and improve pre-training and SFT pipelines for high throughput; Develop and improve inference and evaluation stack including inference engine and pipelines; Build and improve interaction and data flow among RL system roles; Integrate and orchestrate simulation and robotics environments as RL environments; Build and refine distributed training backend with sharding and parallelism; Improve efficiency, scalability, and resiliency of training and RL workloads; Define reliability and efficiency metrics; Root cause, triage, and resolve failures from application to hardware level.
Seniority
Senior, hands-on IC
