CHECK-01
Governance in place?
AIMS, risk processes and responsibilities from stage 1 are established.
AI in your product — when you are ready for it.
Only those who master governance and their own processes can bring AI into their products responsibly. That is the only path that actually works — and the reason this stage is third, not first. The third stage on the way to the AI-first company.
SEQUENCE / 01
Why product AI comes last
An AI feature in your product means: your customers rely on it, your name is on it, the EU AI Act regulates it. Attempting that without a foundation builds the risk straight into the product.
CHECK-02
Processes running?
Your development process uses AI productively (stage 2) — your team knows the technology from daily work.
CHECK-03
Business case clear?
The AI feature solves a real customer problem — not just a line on the data sheet.
Not there yet?
No problem — then this is your starting point: Stage 1: Governance & Readiness or Stage 2: AI in the development process. Both pay directly into the later product AI.
SERVICES / 02
From idea to an auditable AI feature
-
AI architecture & make-or-buy
Which model, which infrastructure, self-host or buy? We design the architecture of your AI feature along your requirements for latency, cost, data protection and maintainability — vendor-independent.
-
Compliance by design
EU AI Act risk class, transparency duties, technical documentation, conformity assessment: we build the regulatory requirements into product design and development process from the start — not retroactively.
-
Safeguarding & testing AI features
AI features are tested differently from deterministic software: data quality, edge cases, robustness, monitoring in the field. As test engineers we design test concepts that an auditor understands too.
-
Operations concept & lifecycle
Models age. We define how your AI feature is monitored, retrained and versioned in the field — including recall scenarios and the update processes your industry demands.
PRINCIPLE / 03
Your product, your data, your AI
Our principle applies in the product too: no black box. You understand what your AI feature does, you control the data, and you do not depend on a single vendor. Where self-hosting makes sense, we build self-hosted — where it does not, with a clean exit scenario.
NEXT STEP / 04
Which AI feature is your market already expecting?
In an initial consultation we test your product idea against feasibility, regulation and your current maturity — honestly, even if the answer is "not yet".
FAQ
Frequently asked questions
FAQ-09Does our data leave the house?
Only if you decide so — and never uncontrolled. Our principle is self-hosted first: sensitive data, engineering knowledge and customer data stay on your infrastructure or in German data centres under your contract. Where cloud AI is the better choice, with a sound legal basis, a data processing agreement, and no training of third-party models with your data.
FAQ-10What does "self-hosted" mean — and do we need it?
Self-hosted means: the AI models run on servers you control — instead of at a US provider. You do not need it everywhere: for uncritical tasks, cloud AI is often more economical. For trade secrets, development data and personal data, control over the infrastructure is the cleanest path. We build both and recommend per use case — not per ideology.
FAQ-12What if the AI does something unforeseen?
Nothing is worse than an AI agent doing unforeseen things — so we do not build any. Our AI processes run deterministically: an unambiguous task, a defined scope, defined gates and clear handovers to humans. No system that finds its own tasks. As engineers we treat it like any machine: behaviour you cannot predict is not a feature, it is a defect.
FAQ-05What does the EU AI Act actually require from us?
That depends on which AI you use and what for. The EU AI Act sorts applications into risk categories — from minimal (hardly any obligations) to high (extensive documentation, transparency and control duties), with staged deadlines and substantial fines. The first step is always an inventory and classification of your AI applications — after that you know exactly which obligations apply to you.