AI Infrastructure Performance Engineer (GPU/LLM)

RadixArk; Palo Alto (CA); On-site; USD 200,000 - 400,000; Full time

Job summary

RadixArk is seeking a Member of Technical Staff — Performance in Palo Alto, CA to push LLM inference and training systems to the limit across real production workloads.

You will tackle latency, throughput, memory efficiency, batching, scheduling, and distributed execution on SGLang, Miles, and the RadixArk stack. This role focuses on measurable production metrics and cost-per-token optimization.

Qualifications

• Strong systems engineering background with performance focus.
• Experience with GPU systems, distributed systems, or ML runtimes.
• Familiar with profiling tools, tracing, and benchmarking methodologies.

Responsibilities

• Analyze and improve performance across RadixArk deployments.
• Benchmark LLM inference/training workloads on GPUs/TPUs/clouds.
• Optimize latency, throughput, memory usage, batching, scheduling, and GPU utilization.
• Investigate performance regressions in real customer environments.
• Collaborate with kernel, runtime, distributed systems, and product engineers.
• Build internal tooling for profiling, tracing, benchmarking, and regression detection.
• Translate workload characteristics into concrete performance tuning strategies.
• Define performance metrics including cost-per-token and serving efficiency.
• Partner with customers and cloud partners on technical evaluations.
• Contribute performance insights back to open-source components.

Skills

Systems engineering

GPU systems

Profiling and debugging

Python

C++

Benchmarking

Tools

CUDA

Triton

ROCm

XLA

Job description

RadixArk is seeking a Member of Technical Staff — Performance in Palo Alto, CA to push LLM inference and training systems to the limit across real production workloads.

You will tackle latency, throughput, memory efficiency, batching, scheduling, and distributed execution on SGLang, Miles, and the RadixArk stack. This role focuses on measurable production metrics and cost-per-token optimization.