AI in critical systems
We make AI-assisted software development fit for critical and regulated systems. There, speed is not the bottleneck. The question is who stands behind the result and can prove it.
What is shifting
What used to take ten consulting days, your team now largely does on its own with Copilot or Claude. What gains value is what an LLM cannot do, or must not do.
Getting cheaper
Getting more valuable
No model knows your organisation, its legacy, its politics and its regulation.
Monolith or microservices, build or buy, cloud or on-prem: an LLM gives you options, but nobody answers for them.
DevOps is largely team structure and collaboration. No prompt changes that.
When your team produces five times more code, it needs more architectural discipline, reviews, tests and security checks, not less.
Our niche
Public sector, critical infrastructure, insurance and automotive have one thing in common: regulation, audits, long-lived systems and liability shape everyday work. That is exactly where AI is most tempting and most risky at the same time. The pressure to modernise is high, skilled people are scarce, but “move fast and break things” is not an option.
What we focus on
Running LLM applications in production: RAG architectures, model gateways, observability for prompts and responses, evaluation pipelines and cost control for inference. On-prem or in an EU cloud if you need it.
Bringing coding agents safely into your pipelines and development process, with clear rules for permissions, sandboxing, reviews, metrics, licences and data protection.
LLMs make migrations such as COBOL to Java or monolith to services economical for the first time. We own the target architecture and steer the AI-assisted migration step by step, with the strangler pattern instead of a big bang.
Translating the EU AI Act, NIS2 and DORA into technical measures: policy as code, traceable pipelines and audit evidence your system generates by itself.
Threat modelling for LLM applications and agents: prompt injection, tool permissions and data exfiltration. A young field with few real experts.
What you can hold us to
We use AI as a matter of course. But every architecture decision and every change to a critical system has a named person who is accountable for it.
What we build produces its own audit evidence: policy as code, traceable pipelines, documented architecture decisions. AI-generated code in particular needs complete traceability.
A system is only done when it can be handled during an incident at three in the morning. That is what sets us apart from pure development shops.
Your legacy systems hold decades of business logic. AI makes it economical for the first time to understand them and modernise them step by step instead of replacing them at great risk.
We use AI tools internally under exactly the governance rules we recommend to you. So our advice is tested in everyday work.
How we start
In two to three weeks we work out where your organisation stands, what is allowed, which risks exist and what a safe adoption path looks like. At a fixed price.
We walk the path with your team, sped up by proven reference architectures, pipeline templates and review checklists.
As a sparring partner for reviews, architecture decision records (ADRs) and new AI initiatives, as part of our Advisory.
Our goal is a team that does good architecture work with AI on its own. Not dependence on us.
Who you work with
If you have run critical systems for years, you see within seconds why an AI-generated pipeline would fail an audit. That experience is our product.
Our senior DevOps engineers have worked for public bodies and critical infrastructure operators for years. Focus: secure supply chains, sovereign operation of AI models on-prem or in EU clouds, observability and automated audit evidence.
Software and systems architecture from automotive, insurance and large Kafka landscapes. Focus: AI-assisted legacy analysis, event-driven architectures as the data backbone for AI applications, and the question where an LLM may decide and where it must not.
Financial computer science with a focus on security: threat modelling for LLM applications and agents, and translating DORA, NIS2 and the AI Act into technical measures.
Case studies
SRE and platform engineering for a highly regulated public authority: Red Hat OpenShift and SUSE Rancher/RKE2, end-to-end observability and infrastructure as code, plus third-level support for the internal DevOps team.
Analysis of infrastructure, architecture and development process at a mid-sized insurance company, with a catalogue of measures, a hybrid target architecture and a clear view of VAIT, DORA and NIS2.
For a vessel data analytics provider, we moved the existing architecture of Elasticsearch and manual services to a scalable streaming architecture with Apache Kafka.
As continuous architects, we connect product management, business architecture, operations and the development team, from technical requirements analysis to decisions.
FAQ
Tell us what your team is planning with AI or already doing. We will tell you honestly where the risks are and what a path looks like that you can answer for.
You'll hear back from us within one working day. The first conversation is free and non-binding.