Abstract
Graphs are a powerful and enduring abstraction for representing and utilizing knowledge. They underpin ontologies, knowledge graphs, and many of the systems we rely on to model complex domains. Yet, modern applications such as similarity search, retrieval-augmented generation (RAG), and agent-based AI are exposing clear limits in how traditional graph technologies are designed and used.
In this talk, I present a research journey motivated by a simple idea: to remain relevant, graphs must evolve alongside the applications they support. Drawing from my work on large-scale and multimodal knowledge graphs, similarity search over graph data, and more recent GraphRAG and hybrid graph–vector systems, I discuss how classical graph models and query paradigms are being extended to meet modern requirements.
The talk highlights key challenges and opportunities at the intersection of graphs, embeddings, and AI-driven systems, including hybrid querying, efficient execution, compression, and distribution. The central message is that graphs are not being replaced, but they must be continuously modernized to serve as a foundation for today’s and tomorrow’s intelligent applications.
