Insights
Notes on AI that has to work.
Practical writing on reliability, security and data protection in AI systems, for projects where getting it wrong is not an option.
- Reliability
Evals before go-live: how to know an AI system is ready
A demo that impresses is not a system you can rely on. How to build an evaluation set, choose metrics that matter to the business, and keep quality from drifting after launch.
- Security
Prompt injection is a permissions problem
Once an AI agent can read emails and use tools, anyone who can send it text can try to steer it. Why better prompts will not fix this, and the architecture that does.
- Data protection
Where your data goes in an AI project, and how to keep it in Switzerland
Every AI system moves data through more places than its users see. A practical map of the data flows, the contracts you need under nDSG and GDPR, and the design choices that keep data under control.
Free, 15 minutes, no obligation
Bring one process. Leave with a plan.
Show us one process that costs your team hours every week. After 15 minutes you will know whether AI can take it over, what a 30-day pilot would deliver and what it would cost.
- Swiss company, Swiss quality
- Data in Switzerland or the EU
- nDSG and GDPR