Inference Performance Engineer, Agent Driven Inference Optimization
Core
Optimize large-scale AI inference performance on NVIDIA GPUs using autonomous agent-driven frameworks to maximize throughput and interactivity.
Role type
Senior Inference Performance Engineer (IC)
Builds
Optimized serving frameworks (TensorRT-LLM, SGLang, vLLM, Dynamo) and GPU kernels for AI inference
Domain
AI Infrastructure / GPU Computing / High-Performance Computing
Deliverable
production ML models
Required skills
GPU workload profiling, CUDA kernel optimization, Python engineering, C++ codebase navigation, experimental methodology, AI inference optimization (batching, KV cache, quantization), parallel processing
Preferred skills
Agentic AI workflow automation, disaggregated serving, MoE inference, NCCL/NIXL/NVSHMEM, Hopper/Blackwell architecture expertise, MLPerf benchmark results
Technologies
TensorRT-LLM, SGLang, vLLM, Dynamo, Nsight Systems, Nsight Compute, CUPTI, PyTorch, CUDA, Hopper, Blackwell
Responsibilities
Design reusable optimization workflows for autonomous AI agents; validate and curate agent-generated experiments; profile GPU workloads using kernel traces and roofline analysis; implement upstream patches and optimized kernels; collaborate with architecture and framework teams to deliver performance wins.
Seniority
Senior, hands-on IC
