Platform Engineer & MLOps at Cloudeo · trainer · CEO of Retake
I build cloud and AI infrastructure with a frugal approach (a lot with little, kept secure) and I take apart hardware that was declared dead until it works again.
- 🌍 Based in Strasbourg
- 🚀 Building Phorge, an open-source, self-hostable cloud provider
- 🧠 Exploring Crossplane and Upjet providers, LLMs on unusual NPUs and GPUs, and GitOps at homelab scale
- ✉️ leo@nonnenmacher-logel.fr · Blog · LinkedIn · Retake
Every line of configuration, automation and infrastructure logic is public. The roadmap is a public GitHub project.
| Repository | What it holds |
|---|---|
| Hangar | GitOps (Flux) for the Kubernetes clusters: manifests, Crossplane compositions, docs |
| Network | Router configuration as code with OpenTofu |
| Ansible | Roles and playbooks for everything that is not Kubernetes |
| phorge-api | The Phorge API: tenants, images, instances, memberships and API keys |
Generated with Upjet, to manage Phorge's building blocks from Kubernetes.
- provider-incus: Incus, from lxc/terraform-provider-incus
- provider-openfga: OpenFGA, from openfga/terraform-provider-openfga
- provider-authentik: authentik
- hailo-10h-llm-compiler: compile your own LLM for the Hailo-10H, from a Hugging Face checkpoint to a self-contained HEF
- hailo-mnist: train a model and deploy it on a Hailo-8L NPU
- benchy-graph: display llama-benchy results as graphs
- GladoCam: real-time face tracking with OpenCV and an ESP8266
More on the blog: LLMs on a Hailo-10H and training an LLM from scratch on a 2020 GPU.
Platform
GitOps and automation
Identity and observability
Code, AI and hardware
Generated daily by lowlighter/metrics from this workflow, nothing external to fail.
