A visual future: GraphRAG and AI observability

What if your two-dimensional observability dashboard looked more like the view from the helm of a Star Trek shuttlecraft?

One observability specialist believes that it's possible near-term with a combination of graph neural networks (GNNs) and knowledge graph retrieval-augmented generation (RAG). According to Matt Young, who founded the Cloud Native Computing Foundation's (CNCF's) Observability Technical Advisory Group in 2020, GNN models capable of visualizing spatial information fed with RAG data from knowledge graphs, or GraphRAG, will make that possible.

"We understand the world we live in spatially," Young said. "Instead of using a console and zooming in and zooming out, and holding up rulers to screens, what if I could stand up from it and put on a jet pack and zoom around my workloads, zoom around my cluster farms, and understand them that way?"

Young said in an interview with Informa TechTarget senior news writer Beth Pariseau on an episode of the IT Ops Query podcast that large language models (LLMs) created by convolutional neural networks, which fuel the current crop of generative and agentic AI apps, are just one relatively narrow application of graph data science.

"GNNs ... supersede ... convolutional neural networks and LLMs," Young said. "The patterns and the predictions that they can help us make are really about ... what's the shape of our data?"

Young said GraphRAG has far-reaching implications for application development, the future of work for site reliability engineers (SREs) and software supply chain security.

Matt Young

App developers and DevOps teams could use AI agents backed by GraphRAG to simulate deployments to infrastructures too large for humans to keep in their heads with spatial visualizations. Similarly, SREs could respond faster during incidents using richer visual signals, and SecOps teams could more quickly ascertain whether apps or infrastructure with complex dependencies contain vulnerabilities.

"If you imagine that every atom in a molecule is a node in one of these graphs, and all of the bonds between those atoms and other atoms are represented by edges -- relationships between them -- we're seeing advances in medicine and material science because they've been able to apply the same kind of predictive analytics," Young said. "Not based on pure statistics, and narrowly looking at a string of numbers ... but that actually models the interactions between components in these complex systems."

Young said he doesn't believe that agentic AI systems will replace human SREs, however.

"I would not say a year or two years from now that the systems we build are going to be less complex or smaller," he said. "The need to understand what's working and what's not, I would expect to expand."

Beth Pariseau, senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.