Retrieval Augmented Generation (RAG) is the cornerstone of a larger pattern called Generative AI (GenAI). That larger pattern includes the use of RAG, a Prompt, and a Large Language Model (LLM) to generate a response.

The GenAI flow first searches previously generated embeddings for data that is semantically similar to a given input, then combines those results with a premade prompt, and submits to an LLM for completion. The response from the LLM offers some insightful (possibly complex) answers that would be unreasonably difficult for a human to attempt.

Experiments on the Vectorize Platform give you an easy way to upload data, generate embeddings, and store it in a vector store. That's the RAG ingredient of GenAI. You can then bring all that work into the RAG Sandbox to see the complete generative AI flow.

While creating and running experiments you chose a certain vectorization strategy which included multiple combinations of embedding models and chunking configurations. Those strategies are also brought into the RAG Sandbox, where you can use the same GenAI flow with those different embedding models. This gives you further insight into the best model and chunking combination for your data.

Last updated