51Your RAG keeps retrieving near-duplicate chunks, wasting the context window. How do you diversify results?▼mediumGleanPerplexityNotion1 replies◆ premiumTop-k by raw similarity often feeds the model five paraphrases of one paragraph, denying it the other facts the question needs. The fix is ranking for relevance and novelty together. Here is how.Open full answer →
46How do you balance relevance and diversity in a ranking/recommendation list?▼hardNetflixSpotifyPinterest1 replies◆ premiumRank items by relevance alone and you ship ten near-identical results and a filter bubble. What interviewers want is a re-ranking stage that scores the list as a set, plus the metric that proves it works.Open full answer →
