Recent advances in Large Language Models (LLMs) have enabled a new generation of intelligent software systems capable of autonomous reasoning, planning, tool usage, and collaboration. Rather than operating as isolated conversational agents, these systems combine multiple specialized agents into coordinated workflows that can solve complex tasks more reliably and efficiently. This course introduces the fundamental concepts of Agentic AI and demonstrates how modern multi-agent systems can be designed using graph-based execution models.

Introduction to Agentic AI

We introduce the core concepts behind Agentic AI, discussing the evolution from traditional LLM applications to autonomous and collaborative agent systems. The course presents the main architectural components of agentic systems, including perception, reasoning, action, memory, planning, and tool integration, emphasizing how these capabilities enable agents to operate in dynamic environments.

Building Agents with LangGraph

We present LangGraph as a framework for designing and implementing LLM-based agents through execution graphs. Participants learn how agent workflows can be represented as graph structures that support state management, conditional execution, iterative reasoning, and coordination among multiple agents. Practical demonstrations illustrate how graph-based architectures improve modularity, scalability, and maintainability.

Reasoning, Memory, Planning, and Tool Use

We discuss the main cognitive capabilities required by modern agentic systems, including reasoning strategies, short- and long-term memory mechanisms, planning techniques, and the integration of external tools and APIs. The course demonstrates how these capabilities can be combined to build autonomous agents capable of solving complex tasks beyond simple question answering.

Hands-on Development of LLM-based Multi-Agent Systems

The course concludes with practical demonstrations illustrating the development of multi-agent applications using LangGraph. Participants explore how specialized agents collaborate, exchange information, and coordinate execution to accomplish complex objectives. The demonstrations emphasize architectural design, reusable workflows, and best practices for developing robust LLM-powered multi-agent systems suitable for real-world applications.

Prof. Julio Cesar dos Reis is an Associate Professor at the Institute of Computing of the University of Campinas (UNICAMP), Brazil. He received his Ph.D. in Computer Science from the University of Paris-Sud, France, and his Habilitation in Computer Science from UNICAMP. Worked as a Research Assistant at the Luxembourg Institute of Science and Technology (LIST) and currently coordinates research and innovation projects in collaboration with industry and international partners. His research interests include Knowledge Graphs, Ontology Engineering, Natural Language Processing, Generative Artificial Intelligence, Agentic AI, Human-AI Interaction, and Digital Technologies in Education. He currently coordinates the Startup Acceleration Center (NAS) at the H.IAAC Hub, an initiative dedicated to transforming scientific research and AI technologies into innovative products and technology-based startups through collaboration between academia and industry. He has served as Local Co-Chair of the 25th International Conference on Artificial Intelligence in Education (AIED 2024) and co-organized workshops co-located with the International Semantic Web Conference (ISWC). General Chair of the 20th edition of the Workshop-School on Agents, Environments, and Applications (WESAAC 2026). Prof. Julio has coordinated research projects funded by industry and public agencies, including Petrobras, FAPESP, and CNPq, and has authored more than 140 scientific publications in international journals and conferences.