Senior System Engineer (Munich, Germany)

Stellenbeschreibung:

About client

Well-funded and fast-growing deep-tech company founded in 2019. We are the biggest Quantum Software company in the EU. They are also one of the 100 most promising companies in AI in the world (according to CB Insights, 2023) with 150+ employees and growing, fully multicultural and international.

Requirements

Systems Programming Expertise: 10+ years of software engineering experience with strong proficiency in Python . You must be comfortable building system agents, APIs, and CLI tools.
Deep Kubernetes Knowledge: You understand K8s internals beyond simple deployment. Experience with Custom Resource Definitions (CRDs), Operators, and the Kubernetes API server architecture.
GPU Ecosystem Experience: Hands‑on experience managing NVIDIA GPU clusters. Familiarity with NVIDIA drivers, CUDA toolkit, and the container runtime (NVIDIA Container Toolkit).
Linux Internals: Deep understanding of the Linux kernel, cgroups, namespaces, and system performance tuning.
Infrastructure as Code: Mastery of declarative infrastructure tools (Terraform, Ansible) but with a focus on provisioning physical hardware rather than just cloud VMs.
Problem Solving: A proven track record of debugging complex distributed systems where the root cause could be code, network, or silicon.

Preferred qualifications

HPC Background: Experience working with traditional supercomputing schedulers (Slurm, PBS) or modern batch schedulers (Volcano, Kueue, Ray).
Bare Metal Provisioning: Experience with tools like Cluster API (CAPI), Metal3, Tinkerbell, Canonical MaaS, or OpenStack Ironic.
High‑Speed Networking: Knowledge of RDMA, InfiniBand, GPUDirect, and how to expose these technologies to containerized workloads.
AI/ML Familiarity: Understanding of how distributed training works (e.g., PyTorch Distributed, Megatron‑LM, DeepSpeed) and the infrastructure requirements of Large Language Models (LLMs).
Observability: Experience building monitoring for hardware health (DCGM) and distributed tracing for long‑running jobs.

Location

Applicants must have legal authorization to work in the country where the position is based.

What you will be doing

Building the Control Plane: Designing and developing the software layer (APIs, Controllers, Agents) that automates the lifecycle of bare‑metal AI infrastructure.
Orchestrating High‑Scale Compute: Architecting scheduling solutions for large‑scale distributed training jobs across massive clusters of GPUs (NVIDIA H200/B200/B300), ensuring efficient bin‑packing and gang scheduling.
Optimizing the Fabric: Tuning the software‑defined networking layer to support low‑latency interconnects (InfiniBand/RDMA/RoCEv2) essential for multi‑node training.
Developing Kubernetes Extensions: Writing custom Kubernetes Operators and CRDs to abstract complex hardware realities (topology awareness, GPU partitioning) into usable interfaces for our Data Scientists.
Hardware‑Level Debugging: Investigating and resolving deep systems issues, ranging from PCIe bus errors and NCCL communication timeouts to kernel panics on bare‑metal nodes.
Defining Standards: Creating the "Golden Image" for AI workloads, managing drivers, firmware, and OS optimizations to squeeze maximum performance out of the hardware.