Created by team RAGnarok on December 03, 2023

Building an RAG app is easy. However, optimizing its different settings, such as chunk size and context, is necessarily not. With RAG2Rich, one can identify the optimal parameters/configurations based on an answer "richness" score. Answer richness, or the "Rich" score, is a composite metric, based on the context relevance, answer relevance, and groundedness measures computed by TruLens. The average richness score is computed by transforming a relevant vector measure into a scalar using a weighted logistic function. Subsequently, the optimal parameters are chosen based on the optimal value of the average richness score. In other words, RAG2Rich offers a scientific approach to optimizing RAG-based systems. RAG2Rich is built with several technologies, such as Vertex AI/PaLM 2, LlamaIndex, Chainlit, Milvus/Zilliz, TruLens, and Cohere. We demonstrate RAG2Rich with a use case from Substation Automation Systems.

Category tags:

"Great use of the technology to find the optimal configuration through the RICH score."


Josh Reini