medium10 pts · first answer countsragchunking

Choosing a chunking strategy

Your RAG system indexes long policy PDFs. Answers often miss key details that sit just before or after a chunk boundary. What is the best first fix?

A.Index each whole PDF as one chunkB.Split along document structure (headings, sections) with modest chunk overlap, and keep section titles as metadataC.Split every 50 characters regardless of wordsD.Increase the temperature so the model guesses the missing details

Getting reliable JSON output →