Robotic hand and human hand reaching toward each other

AI-Powered Software Development

We build artificial intelligence into your software where it saves work: extracting data from documents, finding knowledge and recognising images – transparent, data-minimising and cleanly integrated.
Language modelsDocument processingKnowledge assistantsLocal models

AI as a feature, not an end in itself

Language models, image recognition and semantic search have matured to the point where they deliver real value in everyday business software. But the value does not come from a chat window placed next to existing systems. It comes from AI features that sit right in the middle of the workflow: a form that fills itself in from an uploaded document, or a search that understands what you mean.

We develop such features as part of custom software – in new applications as well as extensions to existing systems. For every task, we decide afresh whether an AI model is the best solution or whether fixed rules are more reliable and less expensive. If it is still unclear where AI should come in for you, we start with AI consulting.

What AI can take on in your software

Extracting data from documents. Invoices, delivery notes or forms are recognised and the relevant data extracted and presented for review, instead of being typed in by hand.

Knowledge assistants. An assistant answers questions exclusively from your own documents and cites the source, so every answer can be verified.

Semantic search. A hybrid search finds content even when it is worded differently from the search term – across languages.

Recognising images. Photos are analysed and tagged with suitable keywords, for example for documentation or inventory purposes.

Translating. Content appears automatically in the user's language, without anyone having to maintain it by hand.

Classifying and pre-filling. Free text is assigned to a category and recognised details are pre-filled – the decision remains with a human.

Data protection built in from the start

We define which data a model gets to see before the first line of code is written. Where possible, we use open-weight models running on our own servers; where a large model is needed, we turn to European providers. This way, confidential details such as credentials can be kept away from the model in a targeted way.

If you wish, we take care of operations as well: models, database and application then run on our infrastructure and are monitored and updated. We describe how this all fits together on our Operations, Hosting & Cloud page.

From practice

In our own product Inventary, a language model recognises objects in photos, translates content into twelve languages and answers guests' questions based solely on the information provided by the hosts – with source references. It uses the open-weight model Gemma, SAM 2.1 for image segmentation and a hybrid search with PostgreSQL and pgvector. Wi-Fi passwords and other credentials never reach the model.

Our project cost calculator uses AI too: a model on our own servers recognises the project type and the stated requirements from a free-text description. The price is then calculated by a fixed set of rules – so the result remains transparent.

How much does an AI solution cost?

According to our cost model, a chatbot for your own website comes to about €7,000–13,000, AI-powered invoice processing €18,000–33,000 and a locally hosted knowledge assistant €39,000–71,000 – each net of VAT, including project management and quality assurance. On top of this come running costs for operations and model usage.

For your specific project, the project cost calculator provides an initial estimate in just a few minutes.

How we bring AI into your software

  1. Clarify the task, data and quality requirements together
  2. Test a prototype with real sample data
  3. Integrate it into your existing systems and workflows
  4. Operations, quality measurement and further development
  1. Clarify the task, data and quality requirements together
  2. Test a prototype with real sample data
  3. Integrate it into your existing systems and workflows
  4. Operations, quality measurement and further development

Frequently asked questions about AI-powered software

Which AI models do you use?

That depends on the task and the data. For many applications, open-weight models such as Gemma, which we run on our own servers, are sufficient. Where a particularly powerful model is needed, we connect European providers. We do not commit to a single vendor, so the model can be swapped later.

Does our data leave the company?

Not necessarily. If the model runs locally, the data stays on servers that we operate for you. With external providers, we define in advance which data may be transferred and consistently keep confidential details such as credentials away from the model.

How do you prevent the AI from giving wrong answers?

Our knowledge assistants answer only from your own content and cite the source, so every answer can be checked. Where results must be exact and reproducible, for example for prices or bookings, a fixed set of rules takes over – the AI then only provides suggestions that a human confirms.

Can AI be integrated into existing software?

Yes. We access AI features via APIs, so they can be added to existing applications. Your users keep working in the interface they know – only some steps now take care of themselves.

How much does AI-powered software cost?

According to our cost model, a website chatbot comes to about €7,000–13,000, AI-powered invoice processing €18,000–33,000 and a locally hosted knowledge assistant €39,000–71,000 net. You can get an individual estimate using our project cost calculator.

Didn't find your question?Get in touch

Describe the task to us – we will tell you honestly whether AI is the right solution and how we would implement it.

Where could AI save work in your software?

Estimate project costs