(no slide links)

(no transcription)

Paco Nathan  
2024-08-08 12:34:00

Graph RAG has become a buzzword term in the tech industry recently, given the popularity of using knowledges graphs to "ground" LLMs with domain-specific facts. This approach improves the overall quality of responses in AI applications, i.e., reduces "hallucination". It also allows for faster data updates and also helps reduce the need (and costs) for fine-tuning models. This talk explores the origins and architecture for Graph RAG, as well as looking into the different variants: for example, the word "graph" in Graph RAG can mean at least six different things. We'll review some of the popular open source libraries and tutorials, and how to handle Graph RAG practices. In particular, what are some of the implications for graph construction and updating practices in an enterprise environment, such as using entity resolution, entity linking, etc., and what are the implications for downstream AI applications? We'll also recommend resources, such as online forums for Graph RAG developers, related conferences, and top recent books for hands-on deep dives into the code.