Senior AI Inference Performance Engineer - vLLM on GPUs

Advanced Micro Devices; Helsinki; On-site; EUR 120,000 - 180,000; Full time

Benefits offered by this job

AMD benefits

Job summary

Advanced Micro Devices in Helsinki, Finland, is seeking a senior AI performance engineer to push inference efficiency on GPUs using vLLM. You will own end-to-end optimization, profiling, and kernel-level improvements across multi-model deployments.

The role requires deep knowledge of vLLM internals, ROCm integration, and end-to-end workload bottleneck diagnosis, with strong Python and C++ skills. Helsinki or Stockholm options are available.

Qualifications

• 5+ years of software development experience in GPU computing, AI systems, or HPC.
• Experience with vLLM internals and ROCm integration.
• Ability to profile end-to-end workloads from user request to GPU kernel.

Responsibilities

• Drive performance optimization end-to-end on vLLM across leading models and configurations.
• Profile, diagnose, and resolve cross-stack performance bottlenecks in vLLM deployments.
• Diagnose kernel-level performance issues and translate findings into actionable optimizations.
• Contribute to customer-facing engagements with measurable uplifts.
• Integrate and optimize custom kernels within vLLM and related frameworks.
• Optimize multi-node distributed inference with scale-out strategies.
• Contribute to shared performance optimization methodology.
• Leverage AI agents to accelerate daily work and define best practices.
• Upstream optimizations into vLLM and adjacent open-source projects.

Skills

vLLM internals

GPU profiling

Python

C++

ROCm

CUDA

SGLang/TensorRT-LLM familiarity

Multi-node distributed inference

Education

Master's or PhD in Computer Science/Computer Engineering/Electrical Engineering

Tools

HIP

CUDA

Triton

CK

Job description

Advanced Micro Devices in Helsinki, Finland, is seeking a senior AI performance engineer to push inference efficiency on GPUs using vLLM. You will own end-to-end optimization, profiling, and kernel-level improvements across multi-model deployments.

The role requires deep knowledge of vLLM internals, ROCm integration, and end-to-end workload bottleneck diagnosis, with strong Python and C++ skills. Helsinki or Stockholm options are available.