Principal Engineer, Model Performance

present of AI; United States; Remoto; USD 140.000 - 210.000; Tempo integral

Resumo da vaga

present of AI seeks a performance-focused engineer to optimize inference across a large catalog and growing GPU fleet. You’ll perform deep hands-on GPU work, kernel tuning, and serving-engine optimization, making judgements on which gains matter most and proving production quality.

You’ll gain direct experience optimizing inference across model types on custom hardware, using AI agents to scale performance and benchmark generation, while leading a small team and still writing code.

Qualificações

• Experience optimizing inference performance in production environments.
• Ability to lead a small engineering team while contributing code.
• Hands-on CUDA and kernel-level optimization expertise.

Responsabilidades

• Own inference performance across models and hardware.
• Perform kernel tuning and profiling analyses to drive speedups.
• Collaborate with engineers on quantization and serving engines.
• Lead a small team while continuing to write code and validate quality.

Conhecimentos

CUDA

Kernel tuning

Profiling

Inference optimization

Ferramentas

Quantization formats

Serving engines

Descrição da vaga

This role owns how fast and how cheaply every model runs across a very large catalogue and a growing GPU fleet. It asks for deep hands on GPU work, kernel tuning and serving engine expertise, plus the judgement to pick which optimizations matter most. A strong candidate has shipped real performance gains in production, not just measured them, and can lead a small team while still writing the code.

How it fits the AI era

The work is itself about making AI cheaper and faster to serve, so AI is the product rather than a side tool. The role also expects AI to speed up the team's own output, through automated kernel search, profiling triage and benchmark generation. What stays human is the technical judgement: deciding where the next gain comes from and proving quality holds.

How it builds AI experience

A candidate would gain direct experience optimizing inference across many model types on custom hardware, and using AI agents to scale performance work beyond headcount. Building that means hands on work with CUDA, major serving engines and quantization formats, plus practice using LLM tools for kernel tuning and profiling analysis.

Worth showing

• Inference engine production work
• Quantization with quality held
• Benchmark and regression systems
• Leading performance engineers

This is our own read, written from the employer's posting and the company's Spotlight. The employer's posting is the source of truth for the role.

About Runware

High-performance inference platform providing serverless and dedicated GPU infrastructure across thousands of machine learning models.