Creating and Connecting a Knowledge Graph to an AI Agent
In the previous two chapters, we discussed the RAG framework in detail. We started with naïve RAG and then saw how we could add different components, replace others, or modify the entire pipeline for our needs. The whole system is extremely flexible, but some concepts remain the same. First, we start with a corpus (or multiple corpora of texts) and conduct embedding of these texts to obtain a database of vectors. Once the user query arrives, we conduct a similarity search on this database of vectors. Regardless of the scope or type of texts, our pipeline is based on the concept of vectorizing these texts in some way and then providing the information contained in the discovered texts to the LLM.
Texts are often full of redundant information, and in the previous chapter, we saw that LLMs are sensitive to the amount of noise in the input. Most people have seen the benefit of creating schematic notes or mind maps. These...
