Remote ML Inference Engineer - Serving & Performance

Yobitel Communications; United States; Remote; USD 160,000 - 230,000; Full time

Job summary

Yobitel Communications is seeking an engineer to own the model serving layer, optimizing latency and throughput for scale. You will manage vLLM, TensorRT-LLM, and Triton deployments, shaping the benchmarking discipline that keeps performance honest.

You will drive workload-specific serving strategies, quantization choices, and rigorous benchmarking to stay ahead in a fast-moving field. This role is remote-friendly and full-time, with a focus on high-performance inference at scale.

Qualifications

• Shipped at least one production inference stack on specialized GPUs (e.g., H100s or A100s).
• Familiar with vLLM, TensorRT-LLM, and Triton in production contexts.
• Strong opinions about speculative decoding and model serving trade-offs.

Responsibilities

• Own the serving stack selection per workload, including batching modes and caching strategies.
• Lead quantization efforts (FP8, AWQ, GPTQ) and build eval harnesses to prove trade-offs.
• Develop and run InferenceBench-style benchmarks to compare our serving against the field.

Skills

Serving-stack design

Quantization

Benchmarking

Job description

Yobitel Communications is seeking an engineer to own the model serving layer, optimizing latency and throughput for scale. You will manage vLLM, TensorRT-LLM, and Triton deployments, shaping the benchmarking discipline that keeps performance honest.

You will drive workload-specific serving strategies, quantization choices, and rigorous benchmarking to stay ahead in a fast-moving field. This role is remote-friendly and full-time, with a focus on high-performance inference at scale.