VirchowBund - AI chat bot
GDPR-compliant. Secure in the healthcare environment.
An AI chatbot in a sensitive healthcare environment? However, the often serious reservations are unfounded in practice. There are already secure solutions for the relevant areas.
The VirchowBund
- The Association of General Practitioners in Germany
- represents around 144,000 general practitioners and specialists
- founded in 1949.
It provides support with practice management and services such as: Sample contracts, legal advice, practice advice, templates, checklists and webinars.
Faster to my answer.
Instead of struggling through navigation and search hits, a question in natural language is now enough. The chatbot takes you straight to the right content - precise, contextualised and without you having to "understand" the website.
The most important advantages can best be seen in the following examples.
AI in the most sensitive environment? It can - if control is also considered.
Especially in the sensitive medical environment, particularly high demands are placed on reliability, data protection and tonality. At the Virchowbund, one thing was therefore clear: AI yes - but only if the knowledge base, answers and operation are clearly controllable. This is exactly how we implemented the chatbot: as a helpful guide in real time, without the typical uncertainty that many associate with "AI answers".
The bot only uses approved, publicly accessible content from the Virchowbund and does not fetch anything from "anywhere" on the internet. We also worked with the customer to fine-tune the content and language (form of address, "we", notes/disclaimers, priorities). Technically, the system runs in compliance with the GDPR on its own virtual server (incl. vector database): this means that data flows, sources and quality remain traceable - and the Virchowbund retains full control.
This is how we built it.
No "hallucinations"
The bot only uses approved content from the Virchowbund (website) plus curated FAQs - nothing from "anywhere" on the net.
Fine-tuned together
Tonality and rules (form of address, "we" form, references/disclaimers, priorities) were agreed together - so that answers are technically appropriate and remain comprehensible.
RAG principle
We rely on RAG (Retrieval Augmented Generation): The system first searches for suitable text passages in the permitted sources and uses them to formulate the answer - instead of "guessing" freely.
Vector database
A vector database ensures that relevant content can be found reliably even with different formulations.
GDPR-compliant
The chatbot system runs on its own virtual server (incl. vector database). This means that the infrastructure, data flows and operation remain traceable and controllable.
own TYPO3-Extension
We have developed our own TYPO3 FAQ extension for knowledge management - including a REST/interface to the chatbot system. The bot is displayed as a widget directly on the TYPO3 website.
