Design and Implement Intelligent Search DP-800 Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A legal research platform stores 10 million document embeddings. Users search for precedents and require exact similarity rankings because a single missing relevant document could change case outcomes. The current ANN index returns results in 200 ms but occasionally misses a true nearest neighbor. The team is willing to accept higher latency for accuracy. What should they do?

   • ASwitch to an exact nearest neighbor (ENN) search by scanning all vectors and computing VECTOR_DISTANCE for every row.
   • BKeep the ANN index but increase the number of probes to improve recall.
   • CRebuild the ANN index with a different distance metric to improve accuracy.
   • DUse a full-text search query instead to ensure all documents are considered.
   Reveal Solution Discussion Select an answer first
2. 17expert · hard

   A team is designing a vector search solution for a dataset of 100 million short text snippets. They need to minimize storage cost while maintaining fast search. They are considering using 256-dimensional embeddings instead of 1536-dimensional embeddings. What is the primary trade-off they should consider?

   • ALower-dimensional vectors reduce storage and index size but may reduce search accuracy.
   • BLower-dimensional vectors are always more accurate because they reduce overfitting.
   • CLower-dimensional vectors increase search speed but require more CPU for distance calculations.
   • DLower-dimensional vectors are incompatible with ANN indexes.
   Reveal Solution Discussion Select an answer first
4. 18expert · hard

   A company wants to implement semantic search over customer feedback. They have a team of data scientists who can generate embeddings, but they are unsure whether to use a pre-trained model or a custom fine-tuned model. The feedback contains industry-specific jargon. What should they consider?

   • AUse both pre-trained and fine-tuned models and average their embeddings.
   • BUse a full-text search instead because it handles jargon better.
   • CUse a custom fine-tuned model because it can better capture industry-specific terminology.
   • DUse a pre-trained model because it is faster to implement and requires no additional training data.
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A job portal searches resumes. Recruiters search for exact skills like 'Python' and also for broader concepts like 'team leadership'. The current full-text search returns exact matches but misses conceptual matches. They want to improve results without losing exact-match capability. What should they implement?

   • AFull-text search with a custom thesaurus to expand synonyms.
   • BHybrid search with full-text and vector search, combined using RRF.
   • CFull-text search with more stop words removed.
   • DVector search only, because it handles both exact and conceptual queries.
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   Which metric measures the fraction of relevant documents that are retrieved by a search?

   • APrecision
   • BRecall
   • CThroughput
   • DLatency
   Reveal Solution Discussion Select an answer first