AI and automation
Recognise, prepare and complete recurring tasks traceably – with a person as the reviewing authority.
- Evaluate documents and requests
- Prepare replies, tasks and postings
- Automate workflows between systems
We connect knowledge, AI and existing software into solutions for operational tasks – from the first conversation to roll-out.
Choose the situation closest to yours.
Our knowledge is stuck in emails and files.
We bring scattered information together and make every statement traceable to its source.
A pilot that connects project messages, plans and reports for one selected workflow into an evidenced project status – with your documents.
Discuss this situation with us →Our team enters the same information several times.
We design an application where information is entered once and is correct everywhere.
An application for one workflow: interaction concept, implementation and interface to your existing system.
Discuss this situation with us →We want to use AI and keep sensitive data in-house.
We define with you which data may leave the company and implement local, external or hybrid model paths.
A pilot with agreed data paths, selected models and verifiable release rules.
Discuss this situation with us →Recognise, prepare and complete recurring tasks traceably – with a person as the reviewing authority.
Applications that fit the workflow and have a clear visual identity – at the desk, on the shop floor and on the move.
Devices and sensors that deliver data or control processes – together with the software that makes them understandable.
Dirk Wilkens leads projects technically. We bring in specialists for AI and data, software development, design or electronics on a project basis. Your contact is Wilkens Engineering – from idea to operation.
We develop AI systems with you for your own workflows. Together we define which knowledge a model may use, which tools it can operate and when your team approves a decision. To do this, we integrate locally run models, commercial AI services or a combination of both into your existing software.
A prompt describes the task. The harness is the environment in which a model completes that task reliably:
You decide what AI should work on, which data may leave the company and where a person approves. The harness makes these decisions binding and traceable in daily operation.
Model access, tool calls and data releases run through a defined layer with roles, check rules and a log. Models stay interchangeable; deterministic logic such as scheduling or quantity calculations stays outside the model.
Compare local, external and hybrid AI for the same task – and withdraw a release.
For us, security and data protection are part of the integration, not an appendix. What this website shows as a simulation and what we only define and build with you in your project are marked separately.
Shown in the demo Every statement stays linked to its source – down to the original sentence.
In your project Who may see which source is taken from your existing permissions, for example from your directory service, project or document system. The demo has no user roles.
Shown in the demo Only what a rule explicitly releases: in the demo, sanitised wording without names or original text – visible character by character.
In your project Which data types may reach which service is defined with you. You review the contract with the provider – data processing agreement, use for training, retention, region – with us before the first hand-over.
In your project Locally run models run on your infrastructure or in a data centre you choose. With external services, location and region depend on the provider and contract; transfers outside the EU need a basis under Chapter V GDPR.
Shown in the demo Release, withdrawal and every hand-over with recipient and scope appear in the simulated history.
In your project You decide who may release, how long inputs, results and logs are kept and when they are deleted. We implement and document it.
Shown in the demo After a withdrawal there is no new hand-over. What the local model can do continues locally; everything else waits visibly. Content already handed over remains visible in the history.
In your project Whether and when the provider deletes content already handed over is governed by the contract. For a service outage we define the same pattern: continue locally or wait visibly – the scheduling logic keeps working without a model.
Official sources as of 19 Nov 2026
This orientation is not legal advice and not a promise of compliance.
Which decision costs your team too much coordination? Prepare your request in a few minutes – if you like, with a short brief in three answers.
Wilkens Engineering · technical lead: Dirk Wilkens · Kiel