Terkeka
01 — When RAG Sounds Right but Retrieves Wrong
A confident answer can still be built on the wrong evidence. Retrieval quality must be evaluated separately from the final generated answer.
Read articleSeries
Retrieval Quality & AI Evaluation
A multi-part engineering investigation into retrieval quality and evaluation for RAG systems.
RAG & Retrieval Engineering
AI Evaluation
Published
02 — Why Answer Accuracy Is Not Retrieval Accuracy
A correct answer does not prove that retrieval worked. This article uses a reproducible experiment to separate retrieval correctness, evidence support, answer correctness, and the overall RAG result.
Read articlePlanned
Not yet published. Titles and scope may change before release.
- 03Building a Labeled Retrieval Evaluation Set
- 04Recall@K, MRR and NDCG Without the Math Fog
- 05The Dangerous Case: Wrong Retrieval, Correct Answer
- 06Put Retrieval Evals in CI/CD