My name is Evgenii Dushkin I'm an AI/Robotics Engineer working across two main areas: Physical AI & Robotics and LLM Infrastructure.
In Physical AI, I work across the full lifecycle of robotic policy development - from simulation and policy training to Sim2Sim and Sim2Real deployment on physical robotic systems. Experienced with multiple simulation stacks, including NVIDIA Isaac Sim / Isaac Lab with PhysX and Newton backends, MJLab with MuJoCo Warp and PyTorch, and MJX with MuJoCo + JAX. Scalable training setups across multi-GPU and multi-node environments using torch.distributed, as well as Ray-based multi-pod cloud training on Kubernetes/OpenShift.
In LLM Infrastructure, I focus on building production-scale model serving and inference solutions across heterogeneous hardware and cloud environments. I use vLLM and llm-d for scalable and distributed inference across GPU, accelerator, and CPU-only platforms, as well as Kubernetes/OpenShift and cloud deployments. Model quantization, quality evaluation, inference benchmarking, and performance optimization to balance model quality, latency, throughput, memory footprint, and hardware utilization.
My focus is on building practical, scalable AI systems that connect model development, distributed infrastructure, simulation, and real-world deployment.
