Skip to main content

Vector-based RAG Embedding

Steps

  1. Overview
  2. Pre-requisites
  3. Key Choices and Guidelines
  4. Implementation

1 Overview​

Description​

Vector databases and Retrieval Augmented Generation (RAG) are pivotal in enhancing the capabilities of AI systems, particularly in the realm of semantic search and natural language processing. Vector databases, such as those supported by SAP HANA Cloud, store data in high-dimensional vector formats, allowing for efficient similarity searches. These databases are crucial for applications like semantic search, recommendations, and anomaly detection. RAG, on the other hand, leverages vector databases to augment the intelligence of large language models (LLMs) by integrating external information retrieval into the response generation process.

In the RAG technique, there are 3 important steps:

  1. Retrieve: Retrieve relevant information from the knowledge base, which is commonly stored in vector databases for faster retrieval.
  2. Augment: Add the additional context fetch in the “Retrieve” step and augment the LLM context.
  3. Generate: Provide a prompt along with the context from the “Augment” step and generate the results using your choice of LLM.

Expected Outcome​

RAG approach helps LLMs to provide more accurate and contextually relevant answers by accessing a vast repository of vector embeddings, which represent semantic similarities rather than mere keyword matches. The combination of vector databases and RAG enables scalable, low-latency semantic search systems, facilitating advanced applications such as open-domain question answering and data-to-text generation. This best practice will create a deeper understanding of the RAG process flow and the creation of embedding in the Vector Database.