Remote Principal Engineer, Model Performance & Inference

Speedrun Talent Network; United States; Remote; USD 250,000 - 350,000; Full time

Benefits offered by this job

Stock options

Remote-first culture

Generous PTO

Parental leave

Company retreats

Job summary

Runware is seeking a Principal Engineer, Model Performance to own how fast and how cheaply every model runs across the fleet. You will lead kernel design, quantization strategies and parallelism optimizations, directing the Model Performance team, collaborating with Sonic Inference Engine, infrastructure, and hardware groups to deploy improvements on real traffic.

You will push for day-one performance of new models, select serving stacks, and build a scalable benchmarking/regression system to

Qualifications

• 10+ years in GPU, HPC or deep learning systems engineering with production inference experience.

Responsibilities

• Own inference performance end to end for the fleet: kernels, serving engines, model-level optimizations and benchmarks across image, video, audio, 3D and LLM workloads.
• Set the model performance roadmap and decide models, hardware and techniques to prioritize based on real traffic and cost.
• Write and tune CUDA kernels, fuse attention and MLP paths, bring up new quantization formats with quality maintained, and implement new parallelism/batching on multi-GPU nodes.
• Build the production deployment pipeline for newly released open models from weights to deployment in days.
• Shape the serving stack (TensorRT-LLM, vLLM, SGLang, diffusion/video pipelines) and contribute upstream where appropriate.
• Own benchmarking and regression systems with hardware-aware measurements integrated into CI.
• Collaborate with infrastructure and hardware teams to address GPU memory, storage and network bottlenecks that impact latency.
• Build and lead a performance team, hire senior engineers, and stay hands-on while guiding direction.
• Leverage AI tools for kernel search, profiling, benchmarking and documentation to scale team output.

Skills

CUDA kernels

GPU architecture

Python

C++

Leadership

Tools

TensorRT-LLM

vLLM

SGLang

Triton

PyTorch internals

Job description

Runware is seeking a Principal Engineer, Model Performance to own how fast and how cheaply every model runs across the fleet. You will lead kernel design, quantization strategies and parallelism optimizations, directing the Model Performance team, collaborating with Sonic Inference Engine, infrastructure, and hardware groups to deploy improvements on real traffic.

You will push for day-one performance of new models, select serving stacks, and build a scalable benchmarking/regression system to