AI Integration
A generic chatbot answers questions nobody asked. We wire language models in where they take work off your hands: an assistant that answers from your documents, an AI feature inside your existing software, classification of incoming requests. And we'll tell you honestly when plain automation is cheaper.
Describe your AI task →Typical tasks
- Categorise incoming requests with AI
- Integrate the ChatGPT or Claude API into existing software
- AI assistant on your own documents (RAG, in several tasks)
Every task runs on your board in the Prozess-Abo: with acceptance criteria, delivered 48 hours after we start it — counted Mon 8 am to Fri 4 pm; over weekends and public holidays the clock stops. The first task is free. Scope per task in the service catalogue →
Example acceptance criteria
When we start a task, we record its acceptance criteria on the card. You measure acceptance against them — if a delivery doesn't meet them, we revise it at no extra cost until it does.
- 1The feature answers the agreed test questions correctly
- 2Without grounding, the assistant hands over instead of guessing
- 3AI disclosure is visible to users
- 4API keys are kept server-side, not in the frontend
What you get
- Answers grounded in your data, not the model's general knowledge
- Clear boundaries: it hands over instead of guessing
- AI disclosure under the EU AI Act (Art. 50) built in
- Model choice made with data protection and running costs in mind
Technologies
OpenAI, Claude, Mistral, RAG, Vektor-DB, TypeScript
Describe your first task — it's free.
Create your workspace and put the task on your board. Before we start, we check that it fits into 48 hours (counted Mon 8 am – Fri 4 pm), and deliver it with acceptance criteria.