Lukas Kubelka

Context

Recent advancements in large language model (LLM) research have demonstrated that these models can perform a wide variety of tasks, with the most fundamental being answering questions. At its core, this involves drawing on knowledge acquired during training to generate textual responses. However, this approach has limitations, such as outdated information and the restriction to passive, text-only output. As LLM capabilities have grown, the concept of “AI agents” has gained renewed attention—particularly through function calling mechanisms introduced by OpenAI and others—which aim to enable models to actively perform tasks in the real world, whether physical or digital. Initiatives like Anthropic’s Model Context Protocol (MCP) further advance this vision by proposing ways for LLMs to access external tools.

Aspects

• Building interoperable Agentic AI applications
• Building secure Agentic AI applications
• Incorporating non-agentic sources into Agentic AI applications
• Low-resource LLMs for Agentic AI

Call

If you are fascinated by the idea of AI agents and the provided or related aspects interest you, please contact me and let's have a chat.

lukas.kubelka@kit.edu