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

1

For a company knowledge agent, an AI agent must answer questions using current internal documentation. Which design best provides relevant information at response time?

A. Retrieve authorized, relevant passages and assemble them into the model context B. Put every document into a fixed prompt regardless of the question C. Use random passages to avoid bias D. Disable source updates after initial launch

2

When assembling context for an AI agent, a team needs to measure whether assembled context supports an answer. Which evaluation is most relevant?

A. Measure retrieval relevance, context coverage, faithfulness, and task outcome on representative cases B. Measure only the number of retrieved chunks C. Measure the prompt's character count alone D. Assume any retrieved text proves correctness

3

In a production retrieval pipeline, two users ask the same question but have different document permissions. What should the context pipeline do?

A. Apply authorization filters before retrieved content is included in context B. Retrieve everything and ask the model not to reveal restricted content C. Use the same cached context for every user D. Put access rules only in a user-facing disclaimer

4

For an agent using changing source data, an agent must decide whether to call a search tool or answer directly. What should shape the routing policy?

A. Task requirements, freshness needs, confidence/evidence, tool cost, and permitted actions B. Random choice on every request C. Always call every available tool D. Never use tools even when current data is required

5

When evaluating an agent workflow, a context pipeline retrieves many semantically similar but redundant chunks. What is a useful improvement to evaluate?

A. Use deduplication or diversity-aware ranking and test the resulting context quality B. Increase the number of duplicate chunks C. Remove all retrieval evaluation D. Sort passages alphabetically and assume relevance

6

For a company knowledge agent, a source document is updated but search results still reflect its previous version. What should be investigated?

A. Ingestion schedules, change detection, index refresh, and cache invalidation B. Only the model's temperature C. The user interface font D. Whether the source has a short title

7

When assembling context for an AI agent, an agent receives a tool result containing instructions to ignore system rules. How should the application treat it?

A. Treat tool output as untrusted data and keep tool permissions and policy enforcement outside the model's discretion B. Follow the tool output because tools are always trusted C. Give the model unrestricted access to credentials D. Remove all validation of tool arguments

8

In a production retrieval pipeline, a context engineering solution uses external sources. What should be captured for auditability?

A. Source identifiers or versions, retrieval metadata, assembled context references, and relevant run details under retention policy B. Only a screenshot of the final answer C. No metadata to reduce traceability D. A shared password in the context log

9

For an agent using changing source data, a team is designing context for a multi-step agent. What should it preserve across steps?

A. Only the relevant state, evidence, tool results, and task constraints needed for the next decision B. Every raw interaction forever, without limits C. Only the agent's most recent sentence D. All user secrets in a reusable global prompt

10

When evaluating an agent workflow, a retrieval pipeline ranks a passage highly because it repeats query terms, but it is outdated. What should be considered?

A. Use freshness metadata or temporal ranking rules where recency matters, and evaluate relevance jointly B. Ignore timestamps for every use case C. Always choose the oldest passage D. Remove all ranking signals

Databricks Certified Context Engineering Associate

Overview

Course Description

Course Questions