Case Study: Deutsche Telekom cuts agent build time from 15 days to 2 with Qdrant
Key results
The challenge
Deutsche Telekom's AI Competence Center needed to deploy AI-powered sales and service assistants across the 10 European countries where it operates. Scaling agents in production surfaced distributed-systems problems: tenancy and memory management across regions, horizontal scaling with shared context, and coordinating non-deterministic agent collaboration.
The solution
The team built LMOS (Language Models Operating System), an open-source multi-agent platform-as-a-service, with Qdrant as the vector database backbone for scalable retrieval and context management. It powers the Frag Magenta OneBOT chatbots and voice bots.
“We knew from the start that we couldn't just deploy RAG, tool calling, and workflows at scale without a platform-first approach.”
AJArun JosephEngineering & Architecture Lead, Deutsche Telekom AI Competence Center
The results, in context
LMOS with Qdrant serves as the backbone for Deutsche Telekom's AI services, processing over 2 million conversations across three countries. The time required to develop a new agent dropped from 15 days to just 2.