Graph-Powered Storyworlds: Using Neo4j To Keep 1M+ Word LitRPG Epics Coherent w/ AI
summary
A Neo4j-backed pipeline extracts entities, stats, quests and timelines from DOCX/Markdown manuscripts into a narrative knowledge graph that tracks continuity across 1M+ words, runs Cypher-based static and causal-chain checks, and feeds agents with targeted context (about 2KB per check) instead of full text. Regex first-pass plus LLM coreference and attribute extraction, human-in-the-loop pruning, incremental vs full re-extract tradeoffs, and token cost/observability are all implemented and used in production.
AI-written summary · about
chapters (13)
- 0:00Introduction to graph-powered story worlds
- 1:13Why LLMs fail at long narratives
- 4:43Why knowledge graphs solve coherence
- 6:08Building the LitRPG knowledge graph
- 8:17Solving the pronoun problem
- 10:27Validating facts with rules
- 12:56Dynamic causal chains in stories
- 15:04AI agents using the graph
- 17:14Hard parts of graph extraction
- 20:05Cost and observability of extraction
- 21:51Handling false positives and corner cases
- 23:37Graph amplifies human judgment
- 24:20Q&A on entity extraction
related talks
23mConnecting the Dots With Context Graphs
16mNarrative Extraction with GenAI: Human-centric Knowledge Graph Creation from Text
27mGraphReader Agentic RAG: Rethinking Long-Context Retrieval Systems
28mReconstructing the Past with Knowledge Graphs: Empowering Historians through Neo4j and GenAI
01h 00mGraph Mastery to AI
12mReal-Time Data Transformation for Knowledge Graph with LLMs
28mAgentic GraphRAG: Multi-Agent Knowledge Graph
21mFrom Spaghetti to Structure: Decoding Complex Codebases with Graphs and GenAI