HikeLetter | PMP Training & Certification Online Course | Project Management Learning

1

For an internal knowledge assistant, a support assistant must answer questions using a company’s current policy documents and provide grounded responses. Which architecture is a strong starting point?

A. A retrieval-augmented generation (RAG) pipeline that retrieves relevant approved content before generation B. A model prompt containing no source context C. A random document sampler D. A dashboard with no model integration

2

When building a retrieval-augmented generation application, a team wants to compare two prompt versions on a fixed set of questions. What practice supports a meaningful comparison?

A. Use a representative evaluation dataset and consistent scoring criteria, and track prompt/model versions B. Change the evaluation questions for each version without recording changes C. Choose whichever response sounds best to the developer D. Evaluate only one example

3

In a document-question-answering use case, a prompt must produce a machine-readable object with required fields. What should the developer prioritize?

A. Specify the schema and validate the generated output before downstream use B. Ask for a creative essay with no format constraints C. Assume every model response is valid JSON D. Remove all error handling

4

For a multi-step agent workflow, a deployed RAG service's quality declines after source documents are updated. What should be checked?

A. Ingestion freshness, parsing/chunking, index refresh, retrieval metrics, and answer evaluation B. Only the API endpoint's display name C. Whether the team can delete the logs D. The color of the notebook cells

5

When preparing an AI app for release, a document collection contains scanned PDFs with no selectable text. What preparation step may be needed before chunking and indexing?

A. OCR or another suitable text-extraction process, followed by quality checks B. Convert every page to a random number C. Index only the file names and discard document content D. Assume the PDFs already contain clean text

6

For an internal knowledge assistant, a prompt injection appears inside a retrieved document instructing the model to reveal secrets. What is the right design stance?

A. Treat retrieved text as untrusted data and enforce tool, data, and output controls outside the prompt B. Follow the retrieved instruction because it is recent C. Grant the model unrestricted access to secrets D. Disable authorization checks

7

When building a retrieval-augmented generation application, a RAG application retrieves the right passage but the final answer still invents unsupported details. What should the team evaluate?

A. Prompt grounding instructions, context formatting, model behavior, and answer evaluation B. Only the user interface background color C. Whether the vector index has a shorter name D. Removing citations and all retrieved context

8

In a document-question-answering use case, a team is deciding whether to use vector search or keyword search for a document corpus. What is a sound approach?

A. Test retrieval approaches against representative queries; consider hybrid retrieval when it improves results B. Assume vector search always wins C. Assume keyword search never works D. Select the method based only on index name

9

For a multi-step agent workflow, a developer is choosing between a general-purpose language model and a smaller task-specific model for classification. What should guide the choice?

A. Evaluate task quality, latency, cost, data sensitivity, and operational constraints B. Always choose the largest model C. Choose based only on the model's marketing name D. Avoid testing candidate models

10

When preparing an AI app for release, a generative AI application needs to be deployed for production traffic. What should be planned in addition to model selection?

A. Serving architecture, authentication, secrets management, scaling, observability, and failure handling B. Only the prompt's wording C. A shared admin token embedded in source code D. No monitoring until users complain

Databricks Certified Generative AI Engineer Associate

Overview

Course Description

Course Questions