Senior AI Inference Modeling Architect

Neurophos; San Mateo (CA); On-site; USD 170,000 - 250,000; Full time

Benefits offered by this job

100% health plan premiums

Unlimited PTO

401(k) matching and stock options

Voluntary benefits

Personalized Benefits

Job summary

Neurophos is hiring a performance engineer to own benchmarking and energy metrics for their optical inference accelerator. You will manage measurement methodology, run end-to-end tests, and compare against competing GPUs and accelerators.

The role emphasizes reproducible results, harness development, and codifying workloads for consistent evaluation. The ideal candidate has 5+ years in GPU performance/mechanics, strong Python and Linux skills, and hands-on experience with Nsight tools, Hugging

Qualifications

• Degree in a relevant engineering or CS field is required.
• 5+ years in GPU performance engineering, benchmarking, or ML system measurement.
• Experience turning papers or models into runnable benchmarks.
• Hands-on roofline/limiter analysis or analytical modeling.
• Proficient with Nsight tools and GPU profiling for HBM vs compute.
• Familiar with Hugging Face/vLLM/TensorRT-LLM stacks and MoE concepts.
• Python for harnesses and plots; comfortable in Linux; cloud GPU ops on major clouds.

Responsibilities

• Own performance and energy metrics across fidelities and hardware.
• Produce workload benchmarks across multiple fidelity levels and align datasets.
• Keep workload definitions constant across models, sequences, and batching.
• Integrate inference workloads from external stacks like Hugging Face and Triton.
• Measure GPUs end-to-end, manage cloud or lab accounts, and run recipes.
• Report TTFT, ITL, tokens/second, and energy per token with instrumentation.
• Document discrepancies with RTL simulations and performance models.
• Maintain an internal benchmark suite with clear separation for external claims.

Skills

GPU performance engineering

Benchmark harnesses

Roofline analysis

Limiter analysis

Performance modeling

Python scripting

Linux

Cloud GPU operations

HPC benchmarking

Education

BS or MS in Computer Engineering, Electrical Engineering, Computer Science, or equivalent practical experience

Tools

NVIDIA Nsight Systems

NVIDIA Nsight Compute

TensorRT-LLM

PyTorch

CUDA

Job description

Neurophos is hiring a performance engineer to own benchmarking and energy metrics for their optical inference accelerator. You will manage measurement methodology, run end-to-end tests, and compare against competing GPUs and accelerators.

The role emphasizes reproducible results, harness development, and codifying workloads for consistent evaluation. The ideal candidate has 5+ years in GPU performance/mechanics, strong Python and Linux skills, and hands-on experience with Nsight tools, Hugging