Optimize visual AI model inference across architectures, algorithms, runtimes, GPU kernels, and distributed systems. Profile workloads, identify compute and memory bottlenecks, improve latency, throughput, memory efficiency, and GPU utilization, and develop benchmarking infrastructure. The role involves techniques such as quantization, sparsity, caching, compilation, attention optimization, custom CUDA or Triton kernels, and multi-GPU execution while collaborating closely with research scientists.

As a Research Engineer, you'll design and implement training pipelines for action-conditioned world models, collaborate with industrial partners, and publish impactful research in generative AI applications.