Databricks-GenAI-Assoc Design Applications Practice Question

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 LLM temperature so the model can explore multiple interpretations of ambiguous follow-up questions.

  Why it's wrong here

  Higher temperature increases randomness and makes responses less predictable and less grounded. For a RAG application that must stay faithful to retrieved documents, randomness undermines factual consistency. Ambiguity in follow-up questions is resolved by conversation history and query rewriting, not by sampling diversity. This setting would degrade both conversation coherence and grounding.

• ✗

  Store all conversation turns in a Delta table and query that table at inference time for each user message.

  Why it's wrong here

  Persisting turns in a Delta table is useful for logging and analytics, but querying it synchronously at inference time adds latency and does not by itself provide the model with context. The model needs the relevant turns included in the prompt, and retrieval needs a rewritten query. Storage alone does not satisfy the conversation or grounding requirements, so this is not a required design element.

• ✗

  Disable retrieval for follow-up questions and rely on the model's parametric knowledge.

  Why it's wrong here

  Disabling retrieval for follow-ups removes the grounding mechanism entirely, so answers would rely on the model's training data and could be outdated or hallucinated. The scenario requires responses grounded in retrieved documents, which applies to follow-up turns as much as initial ones. This approach contradicts the grounding requirement and would reduce answer accuracy.

• ✓

  Rewrite the user query using conversation history before performing vector search retrieval.

  Why this is correct

  In multi-turn conversations, the latest user message often lacks the keywords needed for retrieval because it references prior context. Rewriting the query with conversation history produces a self-contained query that retrieves the correct documents. Without this step, vector search may return irrelevant chunks and the model's answer will not be grounded in the right evidence, breaking the grounding requirement.

• ✓

  Maintain conversation history and include relevant prior turns in the prompt sent to the model.

  Why this is correct

  Multi-turn references depend on the model seeing prior messages. Including relevant conversation history in the prompt lets the model resolve pronouns and follow-up intent, which is essential when a user says 'what about the second option' and expects continuity. Without prior turns, each request is stateless and the model cannot interpret references to earlier messages, so the conversation requirement fails.