Building Agentic and Multi-Agent Systems with LangGraph (Pt. 2)

This talk, presented by Greg Loughnane and Chris Alexiuk, co-founders of AI Makerspace, delves into building agentic and multi-agent systems using LangGraph. It covers key concepts like prompt engineering, fine-tuning, Retrieval Augmented Generation (RAG), and agents, with a particular focus on how LangGraph enables cyclical reasoning loops for more complex LLM applications. The session breaks down the code and concepts needed to build industry-standard agentic and multi-agent systems, exploring how to structure agents into teams with defined roles and responsibilities, similar to human organizations. The discussion also touches upon the utility of multi-agent systems for managing complexity and communicating with non-technical stakeholders, and how they can be thought of as function calling all the way down. The presenters walk through building a hierarchical multi-agent system, including a weather research team and a report writing team, demonstrating how subgraphs can be integrated and visualized using tools like Lang graph Studio IDE and LangSmith for debugging.