AI & Automation
We build language models in where they take work off your team – in customer communication, with receipts and documents, in recurring workflows – and leave them out where a simple rule will do. Everything is built for production: answers with sources, traceable logs and no vendor lock-in to a model provider.
Where you can check it
In our own use
The open-source project Dreamline gives AI assistants a long-term memory (MIT license, source code public). In our own bookkeeping, a Claude model reads receipts and suggests the booking – confirmation is done by hand. Before every tax filing, it checks the figures for anything unusual.
The principle behind it applies to client projects too: the model prepares, a person decides. Where a mistake would be expensive – bookings, contracts or replies to customers – the final confirmation stays with a person
What clients bring us in for
An assistant on your own documents
An assistant using RAG (retrieval-augmented generation) answers questions from your knowledge base and cites the source for every answer – instead of relying on a model's training data that nobody can verify.
Reading receipts, emails and forms
The model reads PDFs, emails and forms, sorts them and turns them into structured data. Anything uncertain goes to a person for confirmation instead of silently into the database.
AI inside your existing systems
The integration sits where the work happens – in the CRM, the ticket system, the bookkeeping – instead of a second island next to it that nobody maintains.
Tech: Claude API, Python, FastAPI, Paperless-ngx
Based in Berlin or Brandenburg? What this service means for businesses in the region, including on-site appointments, is on a separate page: AI automation in Berlin
How we work on it
Local or in a German cloud
Depending on your requirements: local models (for example via Ollama) on your own hardware, or cloud models such as Claude. If data protection requires it, your data does not leave the EU.
Traceable, not a black box
Every request to the model is logged and the cost per request is visible. You can see what the model proposed and what it cost.
No model lock-in
Business logic is separated from the model provider. Switching to another provider or an open-source model is a configuration change, not a rebuild.
Questions about this service
Have a project?
Tell us what it is about. In a first conversation we work out with you whether and how we can help – free and without obligation.
