01 — The Blog
Break,
rebuild,
learn,
experiment.
Kubernetes, Terraform, Ansible, homelab, applied AI… The subjects I dig into here, in long-form articles, use cases and step-by-step tutorials.
-
Hermes Agent · Episode 2 September 14, 2026 · 6 minAutomating a job watch with a self-hosted AI agent and MCP
Part two of the 'Hermes Agent' series: plugging an AI agent into an existing tool through its MCP server, writing the brief for a scheduled task, and what to plan for so it doesn't write junk into your database.
Read more ↗ -
September 5, 2026 · 11 minCapturing an idea by voice on Linux with OpenWhispr, without leaving my IDE
On Ubuntu 24.04 in a Wayland session: OpenWhispr running fully locally (Whisper for transcription, Mistral Nemo for cleanup), a bash script, ydotool and a GNOME shortcut. F10 opens a popup, I dictate my idea, Enter files it into a sticky note — my IDE never loses focus.
Read more ↗ -
Hermes Agent · Episode 1 August 27, 2026 · 16 minSelf-hosting a persistent AI agent: Hermes Agent on your own VM
Part one of the 'Hermes Agent' series: running Hermes Agent on your own VM — Proxmox, VPS or local hypervisor. Docker Compose, secrets, firewall, HTTPS, and the traps that leave a deployment green while nothing works.
Read more ↗ -
From Token to Chatbot · Episode 3 August 24, 2026 · 24 minRandomness Under Control: Temperature, Top-k, Top-p
Part 3 of the series 'From Token to Chatbot'. How an LLM actually picks the next word: greedy, sampling, and the three dials temperature / top-k / top-p, with hands-on Python to feel the balance between reliability and creativity.
Read more ↗ -
From Token to Chatbot · Episode 2 July 16, 2026 · 32 minAttention: How Words Talk to Each Other
Part 2 of the series 'From Token to Chatbot'. The attention mechanism explained with zero math background required: dot product, softmax and the Query/Key/Value trio, with hands-on Python to watch a model link 'it' to 'cat'.
Read more ↗ -
From Token to Chatbot · Episode 1 July 12, 2026 · 36 minHow an LLM Builds a Word
An LLM does only one thing: predict the next word, over and over. Part one of the series 'From Token to Chatbot' — tokens, embeddings, logits and softmax, with hands-on Python to watch GPT-2 hesitate and then pick 'Paris'.
Read more ↗ -
April 25, 2026 · 7 minMy laptop as code: a DevOps approach to setting up my workstation
Why I stopped configuring my laptop by hand. Ansible, mise, uv, direnv, Ansible Vault, pre-commit and a Vagrant smoke test to bootstrap a Debian/Ubuntu workstation reproducibly.
Read more ↗ -
March 31, 2026 · 6 minKandidat — I needed an application tracker, so I designed and deployed it the DevOps way
I needed a tool to structure my job search in Guadeloupe. I designed and built it with Claude Code as support, then automated its deployment on my homelab by putting DevOps culture into practice.
Read more ↗ -
March 30, 2026 · 9 minMy homelab: from Proxmox to Kubernetes, fully automated
How I built an Infrastructure as Code monorepo to manage my entire homelab — Proxmox, Docker, Kubernetes, WireGuard, Pi-hole — with Terraform and Ansible.
Read more ↗ -
February 14, 2026 · 5 minGenerate Excalidraw Diagrams from Claude Code with MCP
How to install and use the excalidraw-render MCP server to generate architecture diagrams directly from Claude Code, in PNG or SVG.
Read more ↗