Talk2024-06-191 min read
Cracking the Code: Deciphering Evaluation Essentials for RAG
Retrieval quality is the single biggest lever on RAG performance. A talk on the evaluation stack every serious RAG team should have.
View original sourceBerlin Buzzwords 2024. Judgment sets, retrieval metrics (Recall@k, nDCG), answer-quality graders, faithfulness, refusal correctness, and how to keep grader drift under control.
The headline: retrieval quality is upstream of everything else in RAG. If you can’t measure it, no amount of prompt engineering will save you.
#rag#evaluation#berlin-buzzwords
