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

Overview

Course Description

Detail	Information
Provider	Anthropic
Exam code	CCAR-P
Duration	120 minutes
Questions	63
Passing score	720 / 1,000 scaled
Exam fee	US$175
Validity	12 months

Course Questions

10 Questions

A retail enterprise wants Claude to answer policy questions from frequently updated internal documents. Which architecture is the strongest starting point?

A. Fine-tune a model on every document and never refresh it B. Place all documents in a system prompt for every request C. Let the model browse an unrestricted public search engine D. Use retrieval-augmented generation with access-controlled retrieval and citations

A team is choosing between a smaller, faster model and a larger, more capable model for a high-volume classification task. What should drive the decision?

A. Always select the largest model available B. Use the cheapest model without evaluation C. Choose based on measured task quality, latency, cost, and risk thresholds D. Choose the model with the longest context window

An assistant must perform a high-impact action such as issuing a refund. Which design is most appropriate?

A. Allow the model to call the payment API with unrestricted credentials B. Trust the model's natural-language statement that the refund is valid C. Require deterministic business-rule validation and appropriate authorization before execution D. Remove audit logging to reduce latency

A system's context window is frequently exceeded by long customer histories. What is the best architectural response?

A. Keep appending all history indefinitely B. Use task-relevant retrieval, structured summaries, and explicit context budgeting C. Truncate the newest user request first D. Increase temperature so the model compresses better

A business workflow has predictable calculations and policy decisions alongside natural-language reasoning. What design is most reliable?

A. Use code or deterministic services for exact logic and Claude for language-heavy reasoning B. Ask Claude to perform every calculation and policy check in free text C. Replace all business rules with a larger prompt D. Avoid testing because the workflow is partly generative

A team proposes a multi-agent architecture for a workflow that one model call already handles reliably. What should the architect do?

A. Compare against a simpler baseline and justify orchestration overhead with measurable gains B. Adopt multiple agents because they are more advanced C. Create an agent for every prompt paragraph D. Skip end-to-end testing

A production assistant needs to distinguish trusted system instructions from retrieved customer content. Which approach is best?

A. Concatenate all content without boundaries B. Allow retrieved documents to override developer instructions C. Disable retrieval D. Maintain explicit instruction/data boundaries and treat retrieved text as untrusted input

A solution must support strict regional data residency. What should be evaluated first?

A. Only the model's benchmark score B. Only the user interface language C. The full data flow, processing locations, service configuration, retention, and contractual commitments D. Whether the prompt contains the region name

A team needs a dependable structured response from Claude for downstream processing. What is the best approach?

A. Ask for JSON and accept any text returned B. Parse values with substring guesses C. Use supported structured-output constraints where available and validate against a schema D. Retry forever until output looks correct

A design must remain robust when the chosen model version changes. What is most important?

A. Hard-code undocumented behavior B. Use model abstraction/configuration, regression evaluations, and controlled rollout C. Disable monitoring after launch D. Assume all model versions behave identically

₹2999.00 ₹999.00

Enroll Now

Claude Certified Architect – Professional