Knowledge Graph Training Program

Everything you need to understand graphs from the ground up: graph theory, knowledge graph fundamentals, the main modeling paradigms, and the technology landscape. More importantly — how graphs, LLMs, and machine learning actually interact, and why that combination is more powerful than any of them alone.

The decisions you make here will haunt you — or reward you — for the rest of the project. Labeled Property Graph (LPG) vs Resource Description Framework (RDF), when to reach for Web Ontology Language (OWL), how to write validation rules with SHACL (Shapes Constraint Language). Get these foundations right and everything else becomes easier.

We run the same real-world dataset through two different paradigms — same questions, different approaches. You'll see exactly how LPG and RDF tackle the same problem, where each shines and where it struggles. Then we explore what your graph can already tell you on its own: centrality, community detection, pathfinding — fast, interpretable, and underrated.

Where graphs actually come from. We build an end-to-end ingestion pipeline — structured data, unstructured text, extraction with open-source LLMs. Then the real question: do you reuse an existing ontology or design your own? We'll work through both paths and understand when each is the right call.

Retrieval-augmented generation is table stakes. Graph-based retrieval is the edge. We build agents that traverse your graph to answer questions no vector store can touch — multi-hop reasoning, context assembly, grounded generation. And yes, we talk honestly about where it still fails.

Stateless agents forget everything after each conversation. Agentic memory changes that — giving your systems the ability to accumulate, update, and reason over knowledge across sessions. We explore memory architectures for knowledge graphs, work hands-on with Cognee, and discuss when persistent memory is a feature versus a liability.

This is where graph embeddings get the depth they deserve. We move from theory to practice using PyTorch Geometric — building models that learn from graph structure, not just node features — applied to real use cases like recommendation systems and fraud detection. You'll leave knowing not just how Graph Neural Networks (GNNs) work, but when to reach for them and when not to.

Building the first version is the easy part. This week is about everything that happens after: monitoring graph quality over time, evaluating retrieval, orchestrating pipelines, and knowing when your system is quietly degrading. The boring stuff that makes or breaks production AI.

The hardest problem in knowledge graphs isn't technical. It's organisational. How do you get a company to think in graphs? How do you build the internal culture, the processes, and the buy-in that turn a proof of concept into a strategic asset? We close with project presentations — each participant presents an end-to-end system built across the nine weeks. Solo or in pairs, formed organically through the Discord community. The conversation most courses never have, ending with something real.