Blog category
Articles on Artificial Intelligence from our project work
Everything devs group has written on Artificial Intelligence, newest first.
An AI system that processes personal data and has to run in Switzerland raises two questions at once, one architectural and one legal. This category treats them together, because in practice they cannot be separated: where personal data actually ends up inside an agent system is decided by the architecture, and what is allowed to end up there by the revised Data Protection Act.
Artificial Intelligence
Sovereign AI inference in Switzerland: the architecture
A provider comparison in CHF, GPU scheduling with DRA, and the points where sovereignty actually breaks. Usually it is not the model.
Artificial Intelligence
A revDSG-compliant AI architecture: what actually counts
The fine of up to CHF 250,000 lands on the person who acted, not the company. What that means for an AI system's architecture, provision by provision.
What this category covers
The articles walk through a revDSG-compliant AI architecture, provision by provision, from the audit log under Art. 4 DPO to the human-in-the-loop gate; and through sovereign AI inference in Switzerland: a provider comparison in Swiss francs, GPU scheduling on Kubernetes, and the points at which sovereignty actually breaks. Both are the most-read groundwork behind our AI practice.
If you have something concrete in mind, from a RAG system to an agent with access to internal systems, the AI agents and LLM integration page describes the offer. In a first conversation we start with the question that decides both budget and architecture: are the data allowed to leave the building, or not?
Every legal reference in these articles has been checked against the primary source, from Fedlex to the ordinance texts, with the date it was checked. We write as architects, not as lawyers: the texts say how to build the requirements, and mark the points where legal advice should take over.
The category sits close to security and compliance, where the reporting obligations are covered, and to AI-assisted engineering, which is about the tooling. Together the three make up the picture we draw for clients in a first conversation: what gets built, what it is allowed to do, and with what.
In projects we turn these articles into systems: RAG setups on our own infrastructure, agents with audit logs and human-in-the-loop gates, inference on Swiss soil where the data demand it. The Husten-Check case study shows such a system in a healthcare setting; the artificial intelligence services page sets the frame for new work.
The projects companies typically bring us: an internal assistant that makes company knowledge searchable without documents travelling to a US provider; an agent that takes over recurring work in an ERP or a support queue, with a log and an approval step; or the question of whether an existing AI pilot is fit to go into production. In all three cases the answer starts with the same sketch — where do personal data flow, where does inference run, who checks the output.
All three routes start with the same no-obligation first conversation through the contact page: fifteen minutes, directly with one of the two founders.