Democratizing data with Graph RAG: What it is, What it can do, How to evaluate it

Democratizing access to data and insights is probably the biggest reason behind the meteoric rise of Generative AI and Large Language Models (LLMs).

LLMs come with both promise and limitations. The most severe and well-known limitation is the fact that they're not trustworthy.

Retrieval-augmented generation (RAG) is an attempt to fix that, and Graph RAG is a variation which has emerged to prominence in the last few months.

So much so, in fact, that Graph RAG is already bigger than Knowledge Graphs in terms of mindshare.

What this means is that there are now many people who are drawn to Graph RAG without knowing much about Knowledge Graphs.

This, and the fact that i have a practical Graph RAG use case, motivated me to document my experience so far.

I go over the background, different Graph RAG architectures, how it works in practice, how to evaluate it, and what's next.

You can read the full article on Orchestrate all the Things.

I hope you will find it useful, and i'm looking forward to your feedback.

Shout-out to those whose work inspired and supported this: Amy Hodler Katariina Kari Paco Nathan Panos Alexopoulos Andreas Blumauer Giuseppe Futia, PhD Heather Hedden Juan Sequeda Ivo Velitchkov Andrea Volpini Jay (JieBing) Yu, PhD Atanas Kiryakov Tomaz Bratanic Ben Lorica 罗瑞卡 Prashanth Rao Morgan Senechal James Phare Mike Dillinger, PhD

To continue the conversation, you can join myself, Jorge Arango , Matteo Casu and Prasad Yalamanchi this Friday in the "Designing Graph RAG Architecture" panel hosted by the SWARM Community

And you can also follow The Year of the Graph and Connected Data