We fix bottlenecks on the shop floor.
With AI that proves its number.
Around 50 engineers, consultants and researchers who put AI where it rarely arrives: into the plant. Built on your real data, measured against a before-and-after figure, hosted in the EU.
Industrial knowledge lives in people, not in systems.
A service technician retires and thirty years of fault patterns leave with them. A quote takes three weeks because the person who can price it is on holiday. A production plan written on Monday no longer holds on Tuesday. None of this is a software gap. It is a knowledge gap that software has never been able to close, because the knowledge was never written down in a form a system could read.
That changed. Language models can read a service report from 2004, a CAD drawing, a WhatsApp thread and a ticket history, and connect them. What they could not do until recently is say where an answer came from. That is the part we build.
“Nobody knew what to do next. The expert had left, and the connections went with them.”
Service lead, machine building, on the situation before the pilotThree product lines, one principle: every answer carries its source.
edda
Reads SAP, SharePoint, Confluence, tickets and email, answers in Teams, Slack or WhatsApp, and cites the document, page and revision behind every answer. Permissions come from the source systems, so nobody sees what they should not.
How edda works → 02 · SOFTWAREAI Engineering & legacy modernisation
Review models for teams shipping AI-generated code, and a way through grown core systems that does not require a big-bang rewrite. Risk classes, guardrails and properties you can actually test.
Read the approach → 03 · PHYSICAL AIRobotics
Cells, cobots and autonomous transport for small batch sizes, with an honest amortisation calculation before anyone signs. Including the cases where the answer is that a robot does not pay off yet.
See the robotics work →Two weeks to a number, then you decide.
Data access, not a workshop marathon
We work with what already exists: service logs, manuals, drawings, ticket history and what the technicians say. Read access is enough. GDPR-compliant, hosted in the EU.
The system learns your estate
Documents get read, linked and indexed with their sources. Answers arrive in the tool the team already uses, so nobody has to open a new tab to get help.
Measurement on real cases
Real faults, real staff, a stopwatch instead of slides. Resolution times and follow-up questions get logged and compared against the previous period.
Your decision
You get the before-and-after figure and the method behind it. If it does not pay off, we say so. Roughly one in five assessments ends with us advising against a build.
TROESTER: six hours to 76 minutes.
TROESTER builds machines and plants in Hanover and has done since 1892. Critical fault knowledge sat with a handful of experienced technicians, in grown folder structures and in decades of service reports. Average time to resolve a fault: about six hours.
What changes across a border, and what does not.
Language and delivery
Our engineering and consulting teams work in English. Documentation, workshops, review sessions and the assistant itself run in English, German or both at once, since most of our clients have German source documents and English-speaking stakeholders.
Hosting and data
Default is EU hosting in Germany. For UK clients we can host in the UK or on your own infrastructure. Nothing leaves your tenant without you deciding it should. Source documents are never used to train a public model.
Time zones
Cologne is UTC+1 or +2, so the UK overlaps almost completely and the US East Coast overlaps by half a day. For US projects we schedule the standing calls in your morning.
Contracts
The pilot is a fixed-scope, fixed-price engagement. It ends with a number and a written method, not an open-ended retainer. What happens afterwards is a separate decision, and a separate contract.
Questions we get asked first
Do you work with companies outside Germany?
Yes. Our client base is concentrated in German-speaking Europe, which is where the manufacturing mid-market we serve is dense. We take projects in the UK, Ireland and the US where the setup fits: an industrial or engineering-led business, a bottleneck that can be measured, and someone internally who owns the outcome.
How much does a pilot cost?
A two-week pilot is a fixed price in the low five figures in euros, depending on how many source systems it has to reach. We quote it after a 30-minute call and a short look at what data exists. If we think the case will not carry the cost, we say that in the call rather than in the invoice.
What if our documents are in German and our team is not?
That is the common case, not the exception. The assistant reads German sources and answers in whatever language the question was asked in, with the citation pointing at the original German document. Multinationals with a German plant and an English head office are a large part of why we built it this way.
Can we run this on our own infrastructure?
Yes. On-premise and private-cloud deployments are supported, including air-gapped setups for sites with no outbound connectivity. This costs more to operate and is worth it for a minority of clients; we will tell you which side of that line you are on.
Who owns what gets built?
You own your data, your documents and the configuration built on them. We keep the product. Nothing from your estate is used to improve a model that another client touches.
Thirty minutes, no sales deck.
You describe the bottleneck, we tell you whether it is worth building, what it would take, and what it would cost. If the answer is that you should not do it yet, you will hear that too.