Agentic Architectural Patterns for Building Multi-Agent Systems
The code repository for a Packt book teaching design patterns for building production multi-agent GenAI systems, covering RAG, fine-tuning, LLMOps and the Agent-to-Agent protocol.
This repository holds the code examples for a book on turning generative AI prototypes into scalable, production-grade agentic systems. It starts from a GenAI maturity model for assessing organizational readiness, then covers model selection, LLM deployment, retrieval-augmented generation, fine-tuning, in-context learning and LLMOps in the context of agentic AI.
It introduces a hierarchical multi-agent architecture in which high-level orchestrator agents delegate sub-processes to specialized agents that collaborate over the Agent-to-Agent protocol, with a practical observability framework using lifecycle callbacks for debugging, compliance and cost tracking. Patterns are illustrated with real-world scenarios and code using the open-source Agent Development Kit, and examples also use LangGraph and CrewAI.
The code is provided as Jupyter notebooks and licensed under MIT.