We explore.We learn.We share.
An honest journal of what we test and what we learn.
We explore.
We learn.
We share.
An honest journal of what we test and what we learn.
11 articles
Featured

Integrating AI in an SME: where to start
Many companies have bought AI licences. Few actually use them. It's not a tool problem. It's an integration problem. Where to start, in what order.

RADIOHED: learning to see on the shoulders of giants
Can you, on a simple laptop, build an AI that reads a chest X-ray and detects pneumonia? The answer holds in one concept: Transfer Learning.

What if I left to raise goats in Chile
Jean-Marc is 50 and has always had WordPress. We propose Next.js. His question: "If you disappear, what do I do?" The honest answer nobody gives.

The machine that doesn't know why it flies
A real dialogue where the AI gets caught flattering, admits it doesn't understand its own functioning, and a dog listening to Bach becomes the best analogy.

AI was already there: you just hadn't noticed
When you searched "pizza Sion" on Google, AI was already auto-completing. YouTube, Spotify, Netflix... AI was everywhere. ChatGPT just gave it a face.

The truth crisis
A video of a president declaring war. A photo of an explosion at the Pentagon. All of it existed. All of it was fake. How do you know what's real?

Between two sets: designing a custom app from real use
An effort rating with 0 entries across 38 sets, a session that could not be logged, a constraint not met. What real use changed in an app built to measure.

The nonsense of AI in marketing in 2026
"AI will revolutionise your marketing." We use these tools every day. Here's what we actually observe, far from the promises.

How an LLM actually works
Everyone talks about ChatGPT. Few people know what GPT stands for, or what happens when you ask a question. It's not magic. It's something else.

Why your website isn't bringing you clients
Most Swiss SMEs we meet have a website that generates no enquiries. No organic traffic, no conversion, no tracking. We see the same mistakes everywhere.

The algorithm that proved it shouldn't be listened to
A stock-analysis algorithm, 174 assets scanned every day. After six months of measurement, it delivered its verdict: don't listen to it. Since then, it has become a coach that would rather refuse to advise than advise wrongly.
Generated posters · art direction by OSOM Labs
"We show results.
Never the recipe.
We document. We share
what actually works."
We document
Real cases. Verifiable figures. No jargon.
We test
Before we talk. Metrics first.
We share
Honest learnings. Mistakes included.
Shall we explore together?
Curious to see how we can help.