AI for sensitive information
Engineering work I do inside a team.
This page describes how I build AI features when the data is sensitive: with a clear, controlled data boundary — on your devices, within your infrastructure, or on trusted Swiss infrastructure. I do this work as an engineer on the team, not as an outside supplier.
- Secure — models are verified, isolated and tested before deployment. Their access is restricted and their outputs are validated before use.
- Private — documents, prompts and results remain within the agreed environment. They are never used to train an external model.
- Regulation-ready — every integration is built around the applicable Swiss data-protection, security and sector requirements, including FINMA expectations for financial institutions.
Choosing the model and the deployment
Picking the model and the hosting approach for the use case is part of the engineering work. That can mean a fully local model, a deployment inside your own infrastructure, or a transparent model such as Apertus on sovereign Swiss infrastructure. Clear data flows and documented controls are written as the system is built, so legal, compliance and security teams have something concrete to review rather than a description after the fact.
If this is the kind of work your team has, my availability explains how I can join it.