@0xMortyx - Google Cloud Workshop: RAG
@0xMortyx on X A 90-minute Google Cloud workshop walks through building production AI agents, moving from basic retrieval-augmented generation (RAG) to Graph RAG, memory, and multi-agent orchestration. The post shares a timestamped chapter guide and links to a written roadmap covering the same progression.
- The workshop is organized as a progression: RAG, then Graph RAG, then memory, then multimodal agent graphs.
- Hands-on sections cover setting up a production agent stack, Graph RAG with semantic search, extracting graph context, and orchestrating specialized agents.
- The author frames it as one of the clearer hands-on examples of teams moving beyond plain RAG toward graph-based context.
The post is a promotional thread built around a 90-minute Google Cloud workshop on moving AI agent systems from basic retrieval-augmented generation (RAG) toward graph-based architectures. The author lays out a progression: RAG, then Graph RAG, then memory, then multimodal agent graphs. A timestamped breakdown maps the workshop's arc: setting up a production agent stack by about 15 minutes in, Graph RAG with semantic search around the 41-minute mark, extracting graph context near 58 minutes, and orchestrating specialized agents from roughly 69 minutes onward. Notably, the thread itself is a tweet quoting an embedded "article" (an X long-form post), and the linked full roadmap sits behind that, so what's visible here is mostly the framing and the video outline rather than the technical content.
The substance worth taking seriously is the core idea. Standard RAG chunks documents, embeds them, and retrieves the nearest matches for a query. That works for lookup but struggles when answers depend on relationships between entities, such as which supplier ships to which plant, or how a policy change propagates through several linked documents. Graph RAG addresses this by representing entities and their connections explicitly, so retrieval can traverse relationships instead of only matching text similarity. The "context graph" framing extends this to agent memory, where past interactions and facts are stored as linked structures the agent can query. The multimodal and multi-agent stages add routing, where a coordinating agent hands subtasks to specialists and assembles their results.
Two caveats matter before you invest the 90 minutes. The thread gives no concrete benchmarks, no named code repositories, and no evidence of how these approaches compare on accuracy, latency, or cost against well-tuned conventional RAG, which is often good enough. Graph construction is also expensive and brittle: entity extraction errors propagate into the graph, and maintaining it as data changes is real work. If you're deciding whether this applies to your problem, check whether your questions are relational in nature. If they're mostly "find the paragraph about X," plain RAG will likely serve you better. The video is worth bookmarking if you're already building retrieval systems and want a hands-on look at the agent orchestration side.