As a scientific collaborator for the iCoSys applied research institute, I am working on complex GPU-related architectures executing machine-learning tasks. Our stack mainly uses Kubernetes and NVIDIA GPUs to provide an easy way to schedule inference workloads, and Slurm to enable experiments and trainings on a queue-based scheduling system. I am the main designer and administrator of our clusters and platforms (Kubernetes, Slurm, vSphere, storage, networks).

Our clusters rely on high-quality hardware and well-defined architectures such as Cisco UCS-X series, A40 and GH200 GPUs, full-10GB network, hybrid and resilient platforms, micro-segmentation, full monitoring…

As a lecturer for future engineers, I teach the Telecommunications and IP Networks course with the following content :

• Introduction to network topologies
• OSI reference model
• Internet Protocol
• Routing and switching
• Medium access (CSMA/CD, TDMA, FDMA)
• Many protocols (ARP, DNS, DHCP, SSH, TCP, UDP, HTTP, RIP, OSPF…)

I am also working as the main designer and administrator of many clusters used by iCoSys institute and whole IT department, described on the experience above.