13 Knowledge graph–powered retrieval-augmented generation
This chapter covers
- Making LLMs useful as AI agents
- Using retrieval augmented generation to ground LLMs using context
- Building a KG-powered RAG system
The year 2023 was marked by AI upheaval. Data scientists and ML engineers working in the natural language processing (NLP) domain were not just given a new toy: their work lives were turned upside down. The release of OpenAI’s GPT-3.5 model at the end of 2022 marked a transformative change. Suddenly we didn’t need to spend months or years building custom training datasets and models for each specific downstream task. With a little clever prompt engineering, almost anyone could build NLP applications.
But although LLMs are powerful, they are far from a magic solution for every problem. A lot of work remains to be done to make them useful in production enterprise-grade scenarios. This realization led to the concept of AI agents, related implementation libraries, and LLM operations (LLMOps).
The topics we’ll discuss in this chapter are illustrated in the mental model in figure 13.1. Parts 2 and 3 of the book showed how to transform private data into a KG; now we’ll explore how to build a chatbot that uses this KG as an input.
Figure 13.1 An AI agent design for KG-powered question answering. The agent has multiple tools at its disposal, which use external data sources such as a vector database and a KG to provide the necessary context for the user’s question.
