Skip to content

Curriculum

14 chapters, 5 parts + a capstone. It moves from concepts (Parts 1–2) through building by hand (Parts 3–4) to evaluation and operations (Part 5).

Every chapter reuses the same corpus and questions from the pharma-mini running project. Keep each part's artifacts and the capstone becomes assembly rather than a restart.

Part 1. Why Knowledge Graphs

Start from the limits of vector RAG, sort out the terms ontology / knowledge graph / taxonomy, and pin down the graph data model (triples).

  1. The Limits of Vector RAG and Where Graphs Come In — where it breaks, what graphs fill
  2. Ontology, Knowledge Graph, Taxonomy — Terms Sorted Out — separating words that get conflated
  3. Triples and Graph Data Models (RDF vs LPG) — the minimal unit for representing knowledge

Part 2. Ontology Modeling

Design a domain as concepts, relations, and constraints, and see the practical path of pulling an ontology out of an existing DB schema.

  1. Ontology Design — Concepts, Relations, Constraints, Hierarchy
  2. Extracting an Ontology from a DB Schema

Part 3. Building a Knowledge Graph (hands-on)

Put data into a graph DB directly, and build a graph from unstructured text with an LLM.

  1. Neo4j and Cypher Basics — nodes, relationships, queries
  2. Building a KG with an LLM — Entity & Relation Extraction
  3. Time-Aware Graphs and Graphiti

Part 4. GraphRAG (hands-on)

Bolt the graph onto RAG. Indexing, retrieval strategies, and a head-to-head with vector RAG.

  1. What GraphRAG Is — Architecture and Indexing
  2. Graph Retrieval Strategies — Local, Global, Community Summaries
  3. Vector RAG vs GraphRAG, Head to Head

Part 5. Evaluation and Production

What to evaluate GraphRAG on, and how to handle graph updates, cost, and latency.

  1. Evaluating GraphRAG — What and How to Measure
  2. Production — Graph Updates, Cost, Latency

Capstone

  1. A Domain Knowledge Graph + GraphRAG Pipeline — Parts 1–5 as one

Progress

This course is released part by part. Start with Part 1.