Improving RAG performance — A Structured Approach (Part 6(A) of RAG Series)
A comprehensive and structured approach
This is part 6 of the “Retrieval-Augmented Generation (RAG) — Basics to Advanced Series”. Links to other blogs in the series are at the bottom of this blog. Taking forward from part 1 (RAG Basics), part 2 (Chunking), part 3 (Embedding) and part 4 (Vector Databases and Vector Libraries) and part 5 (Evaluation of RAG). In this blog, we will focus on one of the most challenging but also the most important step i.e. “Improving RAG performance”
As mentioned in the last blog, developing a basic RAG with all the key components usually does not take more than an hour, but it usually leads to unsatisfactory results. The real challenge, like any other machine learning model, comes around improving the accuracy of the model output and making it production-ready.
i.e. whether the output from the RAG system:
- Is of high-quality content, coherent and factually correct.
- Consistently deliver useful responses
- Is relevant and complete
- Doesn’t have lot of noise
- Is not harmful, malicious and toxic
- Is fast in terms of performance
