Research Topic:

Trustworthy and Scalable Agentic Workflows through Knowledge Grounding and Explainable Decision Backing

Supervisors:

Prof. Michael Färber (TUD), Robert Haase PhD (TUD)

Thesis Abstract:

LLM agents solve complex tasks, but their decisions often lack transparency and break business rules. My research seeks to narrow this gap by combining generative AI with structured knowledge graphs to build trustworthy, scalable workflows. I developed KGNode, a path-aware subgraph extraction method for training-free KGQA. Currently working on derivational backing, providing explicit evidence for generative decisions. Next, I will investigate causal backing, enabling AI agents to justify not only how decisions are derived, but why they satisfy business constraints.