If you are following the trends in AI/LLMs/RAG, chances are that about a year ago you have started noticing a new hot topic - GRAPHS! Proclaimed to be a silver bullet by some, another tool in RAG toolbox by others and overhyped technique by the rest, graphs have certainly made some waves in the LLM community. In this chapter, we get down to the nitty-gritty of graphs - how they actually work, how can you construct them and how can they be used in RAG applications.
Before we start, let me introduce my theory for why the graph boom happened. My take is that graphs are awesome - they represent data in a more logical way, that is suitable for a lot of use-cases. However, couple of years ago constructing graphs was quite hard - you had to extract entities and their relationships by traditional NLP (better case) or by hard-coding rules and using human input (worse case). But now we have LLMs, amazing models that can understand text meaning and nuances (well, until you ask the model to count “r” in “strawberry” 🫢). And that makes the graph creation process WAY easier. So, the final product is RAG based on a graph database, but for me, the main reason for this boom is that we can use an LLM to create the graph database itself.
In this module, we’ll start by introducing the fundamentals of Graph RAG, including what makes graphs a powerful tool for representing and retrieving interconnected data. We’ll explore how graphs help LLMs move beyond isolated facts, enabling contextual and entity-centric retrieval. By understanding the basics of graph databases and entity-relationship networks, you’ll see how Graph RAG structures knowledge in a way that enhances retrieval relevance and response coherence.
Next, we’ll dive into practical aspects, such as constructing and indexing entity-centric graphs. You’ll learn how to transform raw text into structured knowledge using techniques like triplet extraction and node creation, as well as how to optimize these structures for high-relevance retrieval. Real-world applications, such as Microsoft’s GraphRAG for domain adaptation and LinkedIn’s customer support framework, will illustrate the scalability and versatility of this approach across industries.
By the end of this module, you’ll have a comprehensive understanding of how to implement and leverage Graph RAG for cutting-edge RAG applications.