Question

A company is building a RAG-based support assistant over a library of long technical PDF manuals. During testing, the assistant frequently returns answers that mix unrelated topics, and retrieved passages often cut off mid-sentence or combine content from different sections. Which change is MOST likely to improve the relevance and coherence of retrieved context?

Accepted Answer

Adjust the document chunking strategy to respect semantic boundaries such as sections and paragraphs, with appropriate chunk size and overlap

Answer

Increase the foundation model's temperature parameter to encourage more diverse responses

Answer

Fine-tune the foundation model on the technical manuals so it memorizes the content

Answer

Reduce the number of retrieved passages returned to the model to exactly one