Technique
Analysis, methods and insights to drive your digital transformation and master AI.
Custom-Built Solution: What Does It Look Like Under the Hood?
FiscalDoc fits in a few hundred lines of code: a lightweight interface, an engine chosen according to data sensitivity, a simple database, an optional conversational agent. This architecture generalizes to many custom-built projects — until integrations, multiple users, or compliance requirements make it grow.
8 min
How an AI Chooses Its Sources — and How to Become One of Them
Retrieval and query fan-out: two mechanisms, documented by Google since May 2026, decide what a generated answer retains. What gets content into the selection — sharp answers, authority built over time, entity coherence — and what lies beyond anyone's control.
8 min
What does AI already know about your company?
Wikidata, Zefix, professional directories and your own site: four sources that models cross-reference to know who you are before deciding whether to cite you. How to check this identity record in an hour, and correct it at the source — with mcva.ch as a documented example.
8 min
AI Chatbot: Can It Really Serve Your Customers?
The word chatbot covers three systems that have almost nothing in common: scripted decision trees, RAG systems grounded in your documents, agents that take action. What separates a chatbot that serves your customers from one that drives them away is decided at the scoping stage.
9 min
AI project done right: the three patterns that hold up
A well-built AI project is recognizable before the first line of code is written: a pattern suited to the need, carefully prepared data ahead of model choice, a staged progression. The five costliest mistakes — and how to avoid them.
9 min
Claude Code: a field report on two real projects
The redesign of mcva.ch and FiscalDoc were both built with Claude Code. A field report without the marketing gloss: the real gains, the moments where the tool wastes time, and the discipline that changes the outcome.
7 min
Prototyping with AI: what vibe coding is really worth
Describing an idea in natural language and getting a prototype within days: vibe coding delivers on that promise. Keeping it profitable requires a simple rule — validate with the prototype, rebuild for production.
8 min
Structured data: what AI actually reads on your site
JSON-LD markup helps AI systems read your site, but it guarantees no citation whatsoever. The four schemas useful to a Swiss SME, the mistakes that cancel out the effort, and where markup fits within the whole project.
8 min