Demo 1

RAG Chatbot

The assistant at the bottom right of this website answers questions about our services and content. Instead of guessing, it retrieves relevant passages and answers with citations.

How it works

  1. 1Click the button or the chat icon at the bottom right.
  2. 2Accept the privacy notice once.
  3. 3Ask your question – in English or German.

What to expect

Well suited

  • - Questions about services, approach and website content
  • - Answers referencing the underlying source
  • - English and German

Less suited

  • - Current events or real-time data
  • - Confidential or personal input
  • - Topics outside the website context

Test questions to try

Clicking copies the question – paste it into the chat and send.

Security & privacy

  • - Inputs are not stored as training data.
  • - Abuse protection via request and input-length limits.
  • - Answers reference the underlying content (traceable).
  • - Please do not enter confidential or personal data.

Technical details

  • - Retrieval-Augmented Generation (RAG): answers are grounded in the indexed website content, not free model knowledge.
  • - Streaming backend built on FastAPI (Server-Sent Events) – answers appear token by token.
  • - Answer model openai/gpt-oss-120b via OpenRouter (provider Groq), with automatic fallback on outage.
  • - Vector search with Qdrant; frequent requests are accelerated via a Redis cache.
  • - Runs on dedicated infrastructure (Hetzner, EU); protected by rate limiting and input-length limits.