We build AI applications on top of large language models (LLMs): assistants that answer questions from your own documents (retrieval-augmented generation, RAG) and agents that operate tools such as databases, email or your APIs on their own. Depending on requirements we use the OpenAI API, other providers or locally hosted open-source models.
LLM applications that work with your data and complete tasks on their own.
01RAG systems with vector search across documents, wikis, tickets and databases
02AI agents with tool access (function calling, custom APIs, workflows)
03Chat and voice assistants for customers or internal teams
04LLM integration into existing web and mobile applications
05Machine learning models and computer vision (TensorFlow.js, OpenCV)
06Evaluation: answer quality, hallucination checks, cost and latency
03
A look at the code.
This is what an excerpt from a typical project looks like — readable, commented and without magic. That is exactly how we hand code over to your team.
OpenAI APILLMsRAGEmbeddingsPythonTypeScript
rag.pyExample
from qc import VectorStore, LLM
store = VectorStore("handbuecher") # Dokumente → Embeddings
llm = LLM("gpt-4o"if cloud else "local")
def antworte(frage: str, nutzer):
treffer = store.suche(frage, k=6, rechte=nutzer.rollen)
kontext = "\n\n".join(t.text for t in treffer)
antwort = llm.chat(system=REGELN, kontext=kontext, frage=frage)
return antwort, [t.quelle for t in treffer] # immer mit Beleg
04
What you get out of it.
01
Answers with evidence
Every statement points to its source in the document.
02
Permissions preserved
The assistant only sees what the user may see.
03
Swappable model
Cloud today, local tomorrow — without a rebuild.
04
Measurable quality
Automated tests with real questions before every release.
RAG connects a language model to a search over your own data. Before each answer, relevant passages are retrieved and handed to the model. The AI then answers based on your documents rather than general knowledge — traceable and with sources.
How is an AI agent different from a chatbot?
A chatbot answers. An agent acts: it plans steps, calls tools such as databases, calendars or APIs, checks results and completes multi-step tasks. We define exactly what an agent may do and when a human must confirm.
Does our data have to be sent to OpenAI?
No. Depending on the protection level we use cloud APIs under a data processing agreement, EU-hosted models or fully local models on your own hardware.
