Towards Graph-native LLMs and Agentic Systems

Large Language Models (LLMs) continue to advance rapidly, yet their limitations in structured and multi‑hop reasoning highlight the need for graph‑native AI systems. In the first part of this talk, I will present our recent progress on knowledge graph-based retrieval‑augmented generation (KG‑RAG), which improves LLM robustness across data science tasks such as domain‑specific schema matching, entity alignment, and fact verification. In the second part, I will introduce our latest work on agentic LLM systems operating in graph‑structured environments including graphs of tools, actions, memory, and knowledge, and a new evaluation framework that identifies, measures, and explains where misalignment arises during graph‑structured reasoning.