Neo4j is a graph database company that has grown from a small open source project to a team of 800 people over 15 years.
A graph database models data as nodes connected by relationships, which is more efficient than traditional relational database joins for certain queries.
RAG enhances large language models by retrieving relevant external context from a database to augment the prompt before generating an answer.
Graph RAG improves on standard RAG by using a knowledge graph to provide a distilled and highly focused context, reducing noise for the LLM.
Successfully implementing Graph RAG involves cleaning unstructured data and connecting it to business data, starting with a minimum viable graph that evolves over time.
Concerns about data leakage and privacy are driving companies to consider running their own local LLMs for greater control and governance.
The rise of powerful open source LLMs like DeepSeek challenges the dominance of closed source models and changes the financial incentives in the AI industry.
Smaller, specialized language models (SLMs) are enabling powerful, personalized agents that can run locally on devices like phones and watches.
Developers should view agents not as extensions of an LLM, but as a composable software design pattern for controlling and managing LLM capabilities.
While standard RAG uses vector similarity search, Graph RAG excels by connecting disparate pieces of information to provide crucial context that vector search often misses.
Graphs feel unfamiliar to many developers because they are not a native data structure in most programming languages, but pattern matching offers an intuitive way to work with them.
AI tools will increasingly handle boilerplate code, and developers can start exploring graphs and GenAI by taking small, incremental steps without needing to learn everything at once.
29:11Michael Hunger
21:17Carl Lapierre
26:25Ashish Sharma
24:44Tomaz Bratanic
24:42Martin O'Hanlon
59:39Julian Joseph
57:52Rainer Stropek
28:04Phil Nash
Jobs that call for the skills explored in this talk.
