DataEngBytes 2023 - MEL-T2-03 - Joshua Yu
summary
Claims that combining a large language model with a knowledge graph makes generative AI up-to-date, controllable and explainable. Implements embeddings (mentions 1536‑dim vectors), stores them as properties in Neo4j, creates a Neo4j vector index, uses cosine similarity for retrieval-augmented generation, and generates Cypher from natural language to execute against the graph. Pipeline performs semantic search, graph traversal for re‑ranking and entity linking, and a live demo shows sourced, reproducible answers rather than opaque hallucinations.
AI-written summary · about
chapters (13)
- 0:00Generative AI enterprise challenges
- 3:58LLM limitations alone
- 6:52Combining context with large language models
- 8:57In-context learning
- 10:22Retrieval augmented generation workflow
- 11:46Semantic search embeddings
- 14:37Embeddings in knowledge graphs
- 16:25Vector search beyond keywords
- 18:29LLMs in data pipelines
- 19:56Unified data processing
- 20:20Why knowledge graphs enhance large language models
- 22:32Vectors as organizational knowledge
- 23:13Q&A
related talks
01h 17mGraph Database and Retrieval Augmented Generation
31mLive Q&A
38mKnowledge Graphs : Adding a New Dimension to Vector Search and Generative AI
41mGenAI, vector/semantic/hybrid search, RRF, NLP, (...) and other buzzwords
34mEnter the Brave New World of GenAI with Vector Search
19mWisdom-Driven Knowledge Augmented Generation at Scale
30mHow to Build the Next Wave of Graph-Based Agentic AI
17mThe keys to success with Generative AI and Vector Databases