innFactory builds STACKI, a sovereign Llama 3 RAG chatbot for Schwarz Group employee training on STACKIT

Overview

The Schwarz Group needed an internal AI assistant for employee training with the requirement that no data leave Germany. innFactory built STACKI, a RAG chatbot using Llama 3 with STACKIT Model Serving, PGVector in PostgreSQL and STACKIT Kubernetes Engine, hosted exclusively in German data centers.

The challenge

The Schwarz Group uses a sovereign Google Workspace and needed an internal AI assistant for employee training, with the requirement that no data may leave Germany; classic AI solutions such as ChatGPT or Claude were not an option because they run on US servers.

The solution

innFactory developed STACKI, a RAG-based chatbot that runs entirely on STACKIT infrastructure in German data centres. It uses Llama 3 as the open-source LLM with STACKIT Model Serving for GPU-accelerated inference, PGVector in PostgreSQL for vector embeddings of internal training materials, and STACKIT Kubernetes Engine for container orchestration.

Reported business value

According to the source, STACKI runs 100% sovereign in German data centres, searches and summarises internal documents, supports employee training on the sovereign Google Workspace, transfers no data to external providers, and is GDPR compliant and AI Act ready. The source gives no quantified results.