KI-BAS

KI-BAS is a self-hosted, user-facing Retrieval-Augmented Generation (RAG) platform for answering questions about the bidt institute (e.g., people, projects) and its selected digitalization topics. Implemented as a chatbot using state-of-the-art large language models, the platform supports all relevant user questions and requests by retrieving the information most semantically relevant to the query. The full stack runs on local and on-premise infrastructure, and supports routing between local (e.g., Gemma 4) and proprietary models (e.g., Claude Opus 4.7), so sensitive data never needs to leave the organisation.

In addition to providing an on-premise chatbot for data sovereignty and confidentiality needs, KI-BAS demonstrates novel RAG patterns. By combining user queries with the information from the documents, the platform also populates a knowledge graph on digitalization-relevant topics, enabling users to understand the aspects underlying the fuzzy concepts such as „digital platform“ or „software ethics“ more deeply. A built-in rating and evaluation module captures user feedback and logs interaction events, providing a basis for measuring answer quality and studying human-AI interaction. The result is a reusable reference architecture for trustworthy, privacy-preserving generative AI in research and industry.

Bavarian Research Institute for Digital Transformation (bidt)

15.12.2025 – 30.06.2026