RAG needs a map: Neo4j workshop builds GraphRAG agents with vector search plus knowledge graphs
AI Engineer · youtube · 2026-10-11
A hands-on AI Engineer workshop by Nyah Macklin (Neo4j) on using GraphRAG to retrieve connected context that vector search alone misses.
The problem: the five chunks closest to a query can't answer questions requiring traversal of lessons, concepts, or source relationships; a knowledge graph adds connections people can inspect and query.
Build path
- Use an LLM to extract entities and relationships from course material, guided by a schema rather than freeform invention
- Vector index for similarity search; Cypher exposes graph structure; text-to-Cypher enables natural language questions
- Build an agent in an Aura database with a lesson similarity search tool, retrieval queries returning connected graph data, and displayed sources
- A live debugging moment — a lesson count disagreeing with the underlying query — shows why visible evidence matters even after the pipeline works
Code walkthrough covers chunking, entity types, relationship types, and extraction schemas. Resources: GraphAcademy course and the neo4j-graphrag-python library.
More from coding & agent
- Microsoft Foundry demos automatic observability and optimization for enterprise agent fleets — AmyKateNicho · 2026-10-11
- Microsoft demos full agentic AI workflow inside VS Code with Foundry — AmyKateNicho · 2026-10-11
- Veracode: AI now writes half of all code, but security pass rate stuck at 56% — WeldPond · 2026-10-11
- Professor credits Claude with building a bespoke lualatex rig that unblocked a year-stalled book — prof_g · 2026-10-11
- One-month review: Devin SWE-2 + Opus + Codex multi-model coding workflow — CtrlAltDwayne · 2026-10-11
- Building AI agents with real-time web data: Bright Data MCP + Gemini workshop — AI Engineer · 2026-10-11