GPU/ML Engineer (AI)

TL;DR

GPU/ML Engineer (AI): Supporting model optimization and inference for large language models with an accent on NVIDIA GPU architecture and performance tuning. Focus on applying quantization techniques, optimizing workloads for modern GPU hardware, and integrating inference tools to improve AI inference economics.

Company

hirify.global is a recruitment partner focused on connecting engineering talent with specialized technology roles.

What you will do

• Work with NVIDIA GPUs using CUDA to execute and optimize machine learning workloads.
• Apply quantization techniques to LLMs using established libraries like GPTQ.
• Integrate and utilize existing tools for efficient model optimization and inference.
• Optimize model performance specifically for modern GPU architectures such as Hopper and Blackwell.
• Collaborate with the architecture team to validate technical approaches and performance results.
• Rapidly prototype and test technical solutions in production-grade environments.

Requirements

• 5+ years of experience in software engineering, machine learning, or GPU-related roles.
• Strong hands-on experience with NVIDIA GPUs and CUDA programming.
• Proficiency in Python.
• Proven experience working with ML frameworks and running models in production or near-production environments.
• Ability to work independently.
• Solid background in applied mathematics.
• English proficiency is required.

Nice to have

• Experience with LLM optimization and inference pipelines.
• Familiarity with Hopper and Blackwell GPU architectures.
• Knowledge of quantization techniques such as GPTQ.
• Background in embedded systems or low-level optimization.

Culture & Benefits

• Fully remote and flexible engagement model.
• Opportunity for growth into a larger role based on performance.
• Direct engagement with modern AI and LLM optimization challenges.

Hiring process

• Introductory call with the recruiting team.
• Technical interview with the architect.
• Final discussion with company executives if required.