Projects on Stephan Michard

Own Your AI Stack

Mon, 01 Jan 0001 00:00:00 +0000

Turns out “a collection of random thoughts” is not the most compelling way to onboard someone into AI and OpenShift. Who knew. This book pulls together the most useful posts from this blog, reorganizes them into something that actually flows. Same content, less chaos.
If you want to get the latest edition, fill out the form below.

Abstract:
A practical, hands-on guide to building and operating your own AI stack, from lightweight homelab setups to production-ready platforms. This book includes specific code examples and walks through real architectures, covering local AI setups, deploying cloud-based platforms, exposing AI inference endpoints, and working with AI agents.

If you want to move beyond using AI APIs and start owning the full stack, infrastructure, models, and workflows, this guide shows you how to get there step by step.

Built from real-world experiments and field experience, it is designed for engineers and architects who prefer learning by doing. It serves as a practitioner’s companion for hands-on exploration. The examples are intended for learning and proof-of-concept work and are not production-ready; proper consideration of security, safety, and resiliency is essential before real-world deployment.

graphR

Wed, 02 Nov 2022 20:34:39 +0000

graphR grew out of a recurring problem when sizing hyper-converged infrastructure back in the days when I was working at EMC in 2015. RVTools produces a large, detailed Excel export of a VMware environment, covering virtual machines, ESX hosts, and the network configuration. Reading that by hand is slow and easy to get wrong. graphR takes the export, runs the statistical analysis, and assembles the results into a single PDF report with the diagrams you actually need to reason about the environment.

The tool is written in R and ships as a Docker container, so it behaves the same on a laptop or a server. There is also a hosted version at graphr.de for a quick look without installing anything. Uploaded files live in a non-persistent container and are cleared every 15 minutes, so nothing is kept.

The accompanying post walks through how graphR works, how to run the container, and how to adapt the plots to your own reporting layout.

References

Blog Post - link
Web App - link
GitHub Repository - link

Conventional Changelog

Wed, 02 Nov 2022 20:34:39 +0000

Keeping a changelog current is the kind of task that quietly slips once a project gets busy. Conventional Changelog automates it. It reads a repository’s commit history and produces a structured CHANGELOG file, following the Conventional Commits and Semantic Versioning conventions together with the Keep a Changelog format. I added a few commit categories of my own, such as deploy:, gitops:, and demo:, to cover changes that the standard set does not describe well.

The tool fits into whatever workflow is already in place. Run the script directly on your laptop, run it inside a Podman or Docker container, wire it in as a GitHub Action so the changelog updates on every push to main, or add it as a task in a Tekton pipeline on Kubernetes or OpenShift.

A dedicated blog post explains the reasoning behind each convention and provides examples for every execution method. It was also published on opensourcerers.org.

References

Blog Post - link
Opensourcerers Article - link
GitHub Repository - link
GitHub Action on the Marketplace - link

Cloud Bites Tutorial

Wed, 02 Nov 2022 20:34:39 +0000

Cloud Bites is a hands-on tutorial I wrote in 2022, while working at Dell Technologies, to teach the basics of deploying an application in a cloud-native environment. The goal was to educate a community of solutions architects across EMEA about cloud-native technologies, particularly Kubernetes. The chapters build on one another, starting with a container running on your local machine and ending with Kubernetes clusters on Google Cloud. A small website serves as the sample application throughout, and its source code is included in the repository.

Along the way it sets up a reproducible development environment with Vagrant and VirtualBox, builds and runs Docker containers, stands up a local cluster with K3D, works through the everyday kubectl operations for pods and deployments, provisions a cluster on Google Cloud three different ways (the CLI, the Cloud Console, and Terraform), and inspects the running workloads with VMware Octant. It makes no claim to completeness. It is meant as a fast, practical on-ramp for people who learn by doing.

I led several hands-on workshops and curated an online course based on this tutorial. The accompanying blog post provides a chapter-by-chapter walkthrough, and the presentation slides are available below.

Download the slides

References

Blog Post - link
GitHub Repository - link

Compound Interest Calculator

Wed, 02 Nov 2022 20:34:39 +0000

Compound interest is simple as a formula and surprisingly hard to feel. The relationship between how much you save, the rate you earn, and the number of years you stay invested rarely matches intuition. The Compound Interest Calculator is a Shiny app written in R I built to make that relationship visible. Enter a savings rate, an interest assumption, and a time horizon, and it charts how the capital develops year by year.

It goes past a single number. The app compares scenarios, marks milestones such as the year annual interest starts to exceed what you contribute, and breaks the result down across several charts and a detailed table. The idea came from a video about why the first €100,000 is the hardest part of building wealth. I wanted something I could actually experiment with rather than just watch.

You can open the hosted version directly, or run it locally as a container. The full description of every parameter and chart is on the blog.

References

Blog Post - link
Web App - link
GitHub Repository - link

Compound Simulation

Wed, 02 Nov 2022 20:34:39 +0000

Most financial projections tell a tidy, deterministic story: save this much at that rate and you end up with a specific sum in twenty years. Markets do not work that way. Compound Simulation is a Shiny app written in R that takes the probabilistic view instead, running a Monte Carlo simulation over many possible paths to show the spread of outcomes rather than one smooth curve.

You give it the usual inputs: initial capital, monthly contributions, an expected return and a volatility, and a time horizon. It returns a fan chart of trajectories with uncertainty bands, the distribution of end values, and the probability of reaching a target you define. Two extras keep it honest. A stress test halves returns and doubles volatility, and a savings-elasticity view shows whether putting a little more aside each month moves the outcome more than chasing a higher return.

It is the probabilistic companion to the deterministic Compound Interest Calculator . The hosted app is linked below, and the full explanation of the model and its parameters is detailed in the blog post.

References

Blog Post - link
Web App - link
GitHub Repository - link