GraphRAG at scale with Amazon Neptune and Amazon Bedrock

Helpful Links:

• AWS Labs Toolkit
• Try AWS Neptune
• Read the blog: Building a Knowledge Graph
• AWS managed database
• Multi-tenancy with AWS Neptune

In this episode, host Paco Nathan discusses in depth about new GraphRAG capabilities with Ozan Eken and Nicole Moldovan from Amazon Neptune. Their team recently released for general availability (GA) an integration with Amazon Bedrock, so that customers can now create graph vector stores directly from Bedrock Knowledge Bases, for use with Neptune GraphRAG. This leverages Neptune Analytics, a memory-optimized graph database engine for analytics. While the popular Neptune Database services provides a serverless graph database optimized for large scale and high availability, the more recent Neptune Analytics provides an analytics database engine which can ingest large volumes of in-memory graph data, with implementations for popular graph algorithms and support for low-latency analytics queries. Typical use cases include data science workflows, content recommender systems, fraud investigations, detecting network threats, and so on.

For example, in the context of FinCrime use cases, Neptune Database allows for large-scale analysis of fraud tradecraft based on historical data, while Neptune Analytics is suited for attack interventions on the wire. Tying this all together, the new GraphRAG capability with Bedrock provides a fully managed solution that allows users to create GraphRAG applications from their unstructured data (e.g., pdf files, etc.). Users don’t have to be experts on graph technology or graph databases. The graph with embeddings are automatically created from users’ unstructured documents behind the scenes and stored in Neptune. For users who prefer open-source solutions, Neptune also provides an open source Python library which includes the automated construction of a lexical graph from unstructured data sources, along with BYOKG-RAG which is a novel approach for Question & Answer use cases combining structured knowledge graphs with LLMs. 

TL;DR: bring your own structured knowledge graph, integrate large-scale unstructured data sources, then be up and running with a sophisticated Q&A use case in production, with enhanced explainability — all developed and deployed rapidly. The underlying cloud services offer a unique combination of scale, high availability, rapid time to deployment, MLOps observability, offloading of many infrastructure tasks, plus ease of use.

We will talk through some of the lessons learned during these integrations, sharing the pros and cons of combining graph plus vector technologies, and how AWS customers can get started using RAG at scale.