Task Decomposition and RAG as Design Patterns for LLM-Based Systems

AI technologies are moving rapidly from research to production. Compared to traditional AI-based software, systems employing Large Language Models (LLMs) are more difficult to design due to their scale and versatility. This makes it necessary to document best practices, known as design patterns in software engineering, that can be used across LLM-based applications. While Task Decomposition and Retrieval-Augmented Generation (RAG) are well-known techniques, their formalization as design patterns for LLM-based systems benefits AI practitioners. These techniques should be considered not only from a scientific perspective, but also from the standpoint of desired software quality attributes such as safety and modularity. This will help bridge the gap between AI and software engineering as those fine-tuning or prompting LLMs will be aware of the impact that modern techniques have on the overall system.