Most LLM/RAG failures come from missing or inconsistent context rather than model capability. This repository collects tools, agents and samples that demonstrate how to use Google Cloud Knowledge Catalog to create, enrich, and retrieve contextual metadata so downstream LLMs and agents receive semantically organized data rather than raw blobs.
What Sets It Apart
- Practical glue for GCP-based RAG and agent stacks: includes examples that show how to push metadata, build retrieval indices, and fetch contextual snippets for prompts — so you can prototype a retrieval layer on top of Knowledge Catalog without starting from scratch.
- Focus on metadata-driven context management: emphasizes building a dynamic knowledge graph and semantic metadata rather than only file indexing — which means richer entity-aware retrieval and better prompt relevance for agents.
- Samples + agent sketches, not a turnkey product: provides reference implementations and patterns (agents, enrichment pipelines, retrieval adapters) so teams can adapt them into production pipelines on Google Cloud.
Who It's For and Tradeoffs
Great fit if you run workloads on Google Cloud and need structured metadata, entity linking, or a managed knowledge-graph approach to improve LLM inputs. Useful for ML engineers building RAG, AI agents, or governance around data semantics. Look elsewhere if you require cloud-agnostic, production-grade managed services out of the box — the repo demonstrates patterns for Knowledge Catalog and depends on Google Cloud services and SDKs; it's a set of examples and tools under Apache-2.0 rather than a fully managed turnkey product.
