Neo4j’s Role in Fueling Generative AI with Graph Technology

Databases or data structure servers form the backbone of generative AI, powering its capabilities. From real-time champion Redis helping LLM-powered chatbots like ChatGPT to keep the conversation going, to enterprises building chatbots integrating NoSQL hero MongoDB Atlas and Google Cloud Vertex AI PaLM API, database companies are at the forefront of the generative AI revolution, driving its progress. 

Amidst this revolution, graph database pioneer Neo4j takes a strategic approach, playing the long game – i.e. building trust in generative AI.  “We provide fuel for generative AI companies in the form of high-quality, structured graph data,” said Dr Jim Webber, chief data scientist, Neo4j, in an exclusive interview with AIM at their Annual Graph Summit, held last week in Mumbai . 

Neo4j said that it provides the graph data for large language models to be trained on. Over time, knowledge graphs have become vital for organising and accessing enterprise data across industries. Today, Neo4j is playing a pivotal role in helping enterprises integrate LLMs to enhance data handling. They’re focusing on two use cases: developing a natural language interface for knowledge graphs and creating knowledge graphs from unstructured data.

Neo4j’s Graph Data Science framework offers advanced graph analytics and data science capabilities, including a flexible toolkit, a scalable graph database, and a visualisation tool. This enables businesses to make predictions, solve intricate problems, and extract valuable insights. By leveraging graph algorithms and predictive features, organisations demonstrate the value of graphs in advanced analytics, machine learning, and AI. Neo4j’s Cypher is a graph query language resembling SQL to simplify data retrieval from graphs, allowing users to focus on the desired information.

Webber said that the aim is to make data more accessible, understandable, and credible. The integration of LLMs and Neo4j’s graph database technology improves efficiency and accuracy. The project progresses with an open mindset, adapting to new data and technological advancements in this rapidly evolving field. “The convergence of knowledge representation and machine learning is crucial for advancing AI,” said Webber.

Read more: Why Graph Databases Remain Untapped

Dr Jesus Barrasa, head of solutions architecture, Neo4j, told AIM that Neo4j represents the knowledge about a domain, providing curated and explainable answers. On the other hand, machine learning models learn from large datasets and provide plausible but unexplainable answers.

“The combination of these approaches is advantageous,” said Barrasa. He said that machine learning models can distil vast amounts of knowledge, but lack curation and trust. By integrating curated knowledge from Neo4j with machine learning models, we can ask questions about anything with validation and trust. He said that LLMs act as productivity enhancers, allowing quick code generation and data analysis for the knowledge graph population. They also enable non-technical users to interact with technology through generating structured questions from natural language.