Release Engineer (AI)

TL;DR

Release Engineer (AI/DevOps): Designing and managing CI/CD pipelines and hardware-in-the-loop test infrastructure for NPU-based AI accelerators with an accent on version-compatibility and build automation. Focus on automating end-to-end validation, tracking performance regressions, and optimizing the developer feedback loop for compiler and ML frameworks.

Location: Onsite in Seongnam, South Korea

Company

hirify.global is a high-performance AI chip company developing NPUs for next-generation AI workloads.

What you will do

• Own integration CI/CD and manage version-compatibility across PyTorch, vLLM, and RBLN Compiler.
• Build hardware-in-the-loop (HIL) test infrastructure using real NPU boards to automate device allocation and recovery.
• Automate end-to-end validation for numerical correctness and output accuracy across compile, deploy, and inference stages.
• Benchmark core metrics such as latency, throughput, and memory to track and alert on performance regressions.
• Design and manage multi-version build matrices, wheel packaging, and artifact registries.
• Improve developer experience through faster CI, parallelization, and better failure reporting.

Requirements

• Hands-on experience designing and operating CI/CD systems (Buildkite, GitHub Actions, Jenkins, etc.).
• Strong proficiency in Python, shell scripting, and Linux.
• Experience building container-based build/test environments using Docker.
• Experience operating self-hosted runners or custom build infrastructure.
• Familiarity with build systems such as CMake or Bazel.
• Working understanding of ML inference and PyTorch workflows.

Nice to have

• Experience with HIL test infrastructure on accelerators (GPU, NPU, TPU).
• Understanding of vLLM or LLM serving stacks (inference engines, KV cache, batching).
• CI experience on compiler or toolchain projects involving large builds and test matrices.
• Knowledge of PyTorch backend extensions or HuggingFace optimum-style integration.
• Experience with Kubernetes-based runner orchestration or large-scale device-pool management.

Hiring process

• Document review followed by online and on-site interviews, including a technical assignment.
• Culture-fit interview and final compensation negotiation.