Building Evolving AI Agents Via Dynamic Memory Representations Using Temporal Knowledge Graphs

Session Track: AI Engineering

Session Time: 04:15 pm - 04:30 pm November 6

Session description

We live in an age full of AI agents that write emails, summarize documents and meetings, automate workflows, and even provide investment advice. And yet, as paradoxical as it may sound, these systems, despite their increasing presence, tend toward alarming forgetfulness. AI systems typically operate with volatile memory, truncated context windows, and almost no persistent understanding of a user's evolving intentions or environment. This, in turn, leads to repetitions, hallucinations, and a loss of perspective on what is essential. In an age where agents must act autonomously, collaborate with other agents, and interact with users over weeks or months, adaptable memory architectures are not an option but a necessity. This is where temporal knowledge graphs (TKGs) come into play. Unlike static knowledge graphs, TKGs integrate time as a first-class citizen and capture not only what happened, but also when and how relationships evolve over time. This temporal granularity transforms static knowledge into a living, evolving memory, with use cases ranging from personalized recommendations to industrial process monitoring, medical diagnosis assistants, and self-adapting automation workflows. In this session, you will learn how to integrate temporal knowledge graphs based on sophisticated frameworks such as LangGraph and Graphiti on top of a Neo4j-powered graph database as a dynamic memory architecture to build evolving AI systems capable of transforming the way we work, collaborate, and innovate.