AI Inference Performance Engineer — Scale & Speed

Omniva; Bellevue (WA); On-site; USD 200,000 - 300,000; Full time

Job summary

Omniva in Bellevue is hiring a founding performance engineer for Token Factory, a serverless inference platform. You’ll embed with customers, own deployments end to end, and demonstrate gains in tokens per second and time to value during early engagements.

You’ll tune the serving stack (vLLM, SGLang, TensorRT-LLM), optimize models and hardware across NV/AMD fleets, and collaborate with labs. You’ll publish benchmarks and write client-facing engagement reports.

Qualifications

• Senior or Staff level requires 4+ years in performance engineering, ML systems or HPC.

Responsibilities

• Deliver measurable gains for customers from named deployments and tuned production configs.
• Tune the serving system across vLLM, SGLang and TensorRT-LLM on diverse fleets.
• Optimize model and algorithm with FP8/NVFP4/MXFP4 quantization and speculative decoding.
• Tune architecture and hardware with MoE, kernels and advanced attention backends.
• Lead performance agents, review proposals and feed data back to the system.
• Engage with labs and partners to run open-weight releases optimized Day 0.
• Publish reproducible benchmarks and upstream contributions; write engagement reports.

Skills

LLM serving stack

GPU performance fundamentals

LLM inference internals

Customer interaction

Automation

Clear writing

Job description

Omniva in Bellevue is hiring a founding performance engineer for Token Factory, a serverless inference platform. You’ll embed with customers, own deployments end to end, and demonstrate gains in tokens per second and time to value during early engagements.

You’ll tune the serving stack (vLLM, SGLang, TensorRT-LLM), optimize models and hardware across NV/AMD fleets, and collaborate with labs. You’ll publish benchmarks and write client-facing engagement reports.