Databricks Generative AI Engineer Associate Practice Test

Answer choices

1. A. Use model B because its input limit can represent the full chunk, then validate retrieval quality and cost on the target query set.
2. B. Use model A because shorter embedding context always improves semantic precision, for the described technical objective and its associated operational control requirements, within the proposed design.
3. C. Choose either model solely by vector dimensionality because input length does not affect embeddings, for the described technical objective and its associated operational control requirements.
4. D. Use model A and assume the generator's larger context window recovers omitted text, for the described technical objective and its associated operational control requirements.

Wrong-answer review

• B. Use model A because shorter embedding context always improves semantic precision, for the described technical objective and its associated operational control requirements, within the proposed design.: Shorter supported context is not inherently more precise when required content is truncated.
• C. Choose either model solely by vector dimensionality because input length does not affect embeddings, for the described technical objective and its associated operational control requirements.: Embedding dimensionality and maximum input context are separate constraints.
• D. Use model A and assume the generator's larger context window recovers omitted text, for the described technical objective and its associated operational control requirements.: The generator cannot retrieve text that the index vector failed to represent.

Answer choices

1. A. Disable metadata filters so schema changes cannot break them, within the documented operational, security, ownership, and validation requirements.
2. B. Run component tests that create/synchronize the index against the new schema and execute representative filtered queries before promotion, for the required outcome.
3. C. Increase endpoint capacity in production as a preventive measure, for the described technical objective and its associated operational control requirements.
4. D. Run only a spelling check on the prompt because index schemas do not affect retrieval, under organization-wide implementation-governance requirements.

Answer choices

1. A. Embed all comments and choose the nearest response template without classifying intent, as described.
2. B. Use a multi-agent conversation for each comment so agents can debate the category, for the affected environment.
3. C. Classify each comment first; invoke generation only for comments classified as urgent defects, when applied.
4. D. Generate a response for every comment, then classify the generated responses, under the documented operational and governance requirements.

Answer choices

1. A. Total measured cost per successful request/workload, not token count alone, because both quality and latency constraints are already met, for consideration.
2. B. Average answer length only, because concise responses are always cheaper and better, for the described technical objective and its associated operational control requirements.
3. C. Parameter count, because larger models are safer tie-breakers, for the described technical objective and its associated operational control requirements, for this decision.
4. D. Raw token count only, because fewer tokens always means lower cost, under the organization’s defined implementation and exception-management process.

Answer choices

1. A. Record source provenance and license metadata, define allowed uses/attribution obligations by license, and filter ineligible repositories before indexing, for the required outcome.
2. B. Index everything and review licenses only after a customer reports a problem, within the documented operational, security, ownership, and validation requirements.
3. C. Assume all code on a public repository is public domain, for the described technical objective and its associated operational control requirements, for the specified implementation requirement.
4. D. Strip repository metadata so the model cannot reveal where code originated, for the described technical objective and its associated operational control requirements, within this design.

Answer choices

1. A. Add the authoritative service bulletins that contain the missing error-code knowledge, then re-run the failing evaluation set, for the required business outcome.
2. B. Replace the embedding model before changing the source corpus, for the described technical objective and its associated operational control requirements, for this task.
3. C. Increase top-k on the existing manual corpus even though the needed facts are absent, under organization-wide implementation-governance requirements.
4. D. Increase the generator temperature so it can infer likely error meanings, within the documented operational, security, ownership, and validation requirements.

Answer choices

1. A. Add explicit length, CTA, and prohibited-language constraints and include an example that satisfies all three, for the stated application development requirement.
2. B. Increase max_tokens and rely on users to stop reading at 90 words, within the documented scope, ownership, and validation boundaries.
3. C. Lower top-k in Vector Search so the model has less information to discuss, within the stated policy framework.
4. D. Use a larger context window so the model can understand brevity better, as the recommended response to this scenario.

Answer choices

1. A. Grant management permission to every caller because query access does not support authentication, under the organization’s defined implementation and exception-management process.
2. B. Make the endpoint public and enforce authorization only in the prompt, within the documented operational, security, ownership, and validation requirements.
3. C. Use a single shared administrator token in the client application, for the described technical objective and its associated operational control requirements, within the described context.
4. D. Grant the minimum query/use permission needed to invoke the endpoint and reserve management permissions for the operators who administer it, under the documented operational and governance requirements.

Answer choices

1. A. An additional embedding job that rewrites every source document at request time, within the defined security and accountability boundaries.
2. B. A second vector index containing the same vectors but no metadata, for the described technical objective and its associated operational control requirements.
3. C. A post-generation judge that changes the order of citations after the answer is written, for the affected environment.
4. D. A reranker that scores the retrieved candidates and selects the strongest passages before prompt construction, within the stated policy framework.

Answer choices

1. A. Average response length, because unsupported answers are always longer, as the primary implementation for the described business requirement.
2. B. CPU utilization only, because quality drift originates from resource pressure, for the stated evaluation and monitoring requirement.
3. C. A groundedness/correctness quality scorer on sampled production traces, alongside retrieval-quality signals for the changed corpus, within the proposed design.
4. D. Number of prompt versions, because more versions guarantees higher hallucination risk, as the recommended implementation across the complete governed service lifecycle.