Performance Engineer, Containers & Serverless ML on GPUs

Verda; Helsinki; Hybrid; EUR 90,000 - 150,000; Full time

Benefits offered by this job

Cash + equity

Fringe benefits

Job summary

Verda is hiring a Performance Engineer to own the surface for container and serverless GPU platforms. You will profile and optimize the end-to-end path for ML/AI workloads, from image distribution to inference hot path, and drive platform improvements.

You will work across compute, networking, and storage to remove bottlenecks and publish insights to the organization. Helsinki-based, hybrid arrangement with a focus on real-world workloads.

Qualifications

• Production experience improving ML workloads.
• Strong grasp of Linux storage and container internals.
• Hands-on with distributed or network filesystems and S3-compatible storage.
• Familiar with model-serving runtimes and data formats.
• End-to-end view across NIC, switch, filesystem, cache, container runtime and GPU.

Responsibilities

• Profile and optimize end-to-end path for containerized ML workloads.
• Design and tune storage layer between object stores and GPU nodes.
• Drive wins on latency and throughput across services and workloads.
• Benchmark real workloads and translate findings into platform changes.
• Collaborate across compute, networking, and platform teams.
• Publish internal write-ups to explain trade-offs.
• Stay current with evolving ML/AI ecosystem.

Skills

Performance optimization

Linux storage internals

Distributed/network filesystem

S3-compatible storage

Model-serving runtimes

End-to-end systems thinking

Tools

Go

Rust

Python

Docker

Job description

Verda is hiring a Performance Engineer to own the surface for container and serverless GPU platforms. You will profile and optimize the end-to-end path for ML/AI workloads, from image distribution to inference hot path, and drive platform improvements.

You will work across compute, networking, and storage to remove bottlenecks and publish insights to the organization. Helsinki-based, hybrid arrangement with a focus on real-world workloads.