NCA-GENL Software Development Practice Question
A developer is debugging a RAG service where answers are correct in testing but degrade in production as the document corpus grows. Logs show retrieval returning chunks with high similarity scores that do not contain the answer. Which change most directly addresses the root cause?
⚠ Common exam trap
The trap here is reading a high similarity score as evidence of relevance, when dense similarity can be high for passages that share topic but not the answer.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Replace dense embedding retrieval with a hybrid approach that combines dense vectors and keyword matching, then rerank the merged candidates.
Confident but non-answering chunks point to dense retrieval optimizing topical similarity instead of matching the specific terms that carry the answer. Hybrid retrieval adds lexical matching for exact names, numbers, and rare tokens, and reranking the merged candidates pushes truly relevant chunks to the top. This improves precision where it matters, at the head of the retrieved list passed to the model.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Increase the embedding model's output dimension to improve similarity precision.
Why it's wrong here
Higher dimensionality can capture more nuance but does not change the fundamental behavior of dense similarity, which favors broad topical overlap over exact term matches. The failing queries need lexical grounding for names, numbers, or rare tokens. Re-embedding with larger vectors also requires reindexing the corpus and does not by itself fix ranking of confident but non-answering chunks.
- ✗
Increase the number of retrieved chunks passed to the LLM from five to twenty.
Why it's wrong here
Returning more chunks adds more high-scoring but irrelevant passages, increasing the chance the model latches onto misleading context and raising token cost. The problem is retrieval precision, not recall volume. Passing more candidates without improving ranking quality makes the noise problem worse rather than fixing the selection of supporting evidence.
- ✓
Replace dense embedding retrieval with a hybrid approach that combines dense vectors and keyword matching, then rerank the merged candidates.
Why this is correct
High similarity scores without the answer indicate dense embeddings are matching topical similarity rather than the specific terms that carry the answer. Hybrid retrieval adds lexical matching so exact names, numbers, and rare tokens are captured, and a reranker reorders the merged set by relevance to the actual query. This directly improves precision at the top of the retrieved list, which is what the failure mode requires.
- ✗
Lower the similarity threshold so fewer chunks are returned to the model.
Why it's wrong here
A threshold filters by score, but the failing chunks already score highly, so tightening the cutoff may discard relevant passages while still admitting confident but irrelevant ones. The retrieval signal itself is the problem, not the cutoff value. Adjusting the threshold does not add the lexical matching that would surface chunks containing the exact answer terms.
About these practice questions
This NCA-GENL question is part of Courseiva's 367-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official NVIDIA exam blueprint
This NCA-GENL practice question is part of Courseiva's free NVIDIA certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the NCA-GENL exam.