NCP-AAI
Free NVIDIA-Certified Professional: Agentic AI Practice Test
12 exam-style questions with full explanations — no sign-up. Score yourself, then close your gaps with the full course.
Exam fee ~$2006 exam domainsLevel Advanced2 timed practice tests in the course
✅ Free practice — no sign-up📝 Real exam-style questions💡 Detailed explanations💸 30-day money-back via Udemy
An agent must answer a question requiring current stock prices. It needs to alternate between generating a thought, invoking a price-lookup tool, observing the result, and refining its next thought before producing a final answer. Which reasoning paradigm is being implemented?
ReAct explicitly interleaves reasoning traces (thoughts) with actions (tool calls) and observations in a loop, which matches the described think-act-observe cycle.
An insurance company is building a claims-processing agentic system. A central agent receives each claim, determines which specialized agents are needed (document verification, fraud detection, payout calculation), sends work to them, then combines their outputs into a final decision. Which orchestration pattern does this describe?
A single coordinating agent that inspects the task, routes work to specialized workers, and aggregates their outputs into a final decision is the defining characteristic of the supervisor/worker pattern.
An agent orchestrated with NVIDIA NeMo Agent toolkit must complete a 40-step data migration workflow. After step 25, the agent begins optimizing for a sub-task that contradicts the original migration goal stated in the initial prompt. Which architectural change BEST prevents this drift?
Long-horizon agents need an explicit, externally tracked plan/goal-state (not just implicit context) that is checked at each step so sub-goal execution is validated against the original objective, preventing goal drift over many steps.
An agent orchestrates payments through a third-party billing API tool. During a network blip, the agent times out waiting for a response and automatically retries the same tool call. The billing API has no built-in duplicate detection. Which design change BEST prevents a duplicate charge from being processed?
Idempotency keys let the downstream service (or an idempotency-aware wrapper) recognize that a retried request is a repeat of an in-flight or completed operation, so it can safely return the original result instead of re-executing the side effect. This is the standard pattern for making non-idempotent APIs safe to retry in agentic tool execution.
A team is deploying an open-weights LLM for a customer-support agent that must sustain sub-200ms per-token latency under high concurrent load. Which deployment approach BEST meets this requirement?
NIM microservices package TensorRT-LLM-optimized engines behind a Triton-backed serving layer with in-flight (continuous) batching, which is the supported path to hit low-latency, high-concurrency LLM SLAs on NVIDIA infrastructure.
A multi-agent system autonomously initiates financial trades based on market analysis. The risk team requires that no trade above $50,000 executes without explicit human authorization, while smaller trades can proceed autonomously. Which control BEST meets this requirement?
High-risk, high-value autonomous actions require a human-in-the-loop approval checkpoint tied to the actual tool/action invocation (an execution-level control), not a text filter. This ensures the agent cannot complete the action until a human authorizes it, directly enforcing the policy threshold.
A team is automating a multi-step data migration workflow where all steps and their order are known in advance, but any individual step can fail and require the entire remaining plan to be recomputed with the new state. Which agent control pattern should they implement?
Plan-and-execute agents separate planning from execution, generate a multi-step plan upfront, and support re-planning when a step's outcome diverges from the plan — exactly the failure-and-recompute behavior described.
A workflow requires conditional branching, multiple retry loops around failing sub-tasks, and the ability to persist and resume execution state after a crash. Which orchestration framework is BEST suited to implement this as an explicit stateful graph?
LangGraph models workflows as an explicit graph of nodes and conditional edges, supports cycles for retries, and provides a checkpointer to persist and resume state across runs, matching all stated requirements.
In an agentic memory architecture, which statement correctly distinguishes episodic memory from semantic memory?
Episodic memory captures specific, time-bound events (e.g., 'user reported issue X on date Y'), whereas semantic memory holds generalized facts and relationships abstracted away from any single event -- a core distinction tested for agent memory design.
A tool exposed to an agent allows arbitrary Python code execution to help with data analysis tasks. Which implementation BEST satisfies safe tool execution requirements for this capability?
Sandboxed, resource-limited, network-restricted ephemeral containers isolate arbitrary code execution from the host environment, agent process, and sensitive credentials, containing the blast radius of malicious or buggy generated code — the core principle of safe tool execution for code-running tools.
Which NVIDIA component is specifically responsible for compiling and optimizing an LLM's transformer layers — fusing kernels and applying quantization — for maximum GPU inference throughput?
TensorRT-LLM is the library purpose-built to optimize transformer inference through kernel fusion, quantization (e.g., FP8/INT8), and in-flight batching to maximize GPU throughput.
Before releasing an agentic system to production, a security team wants to automatically probe the deployed LLM and its tool-use pipeline for susceptibility to prompt injection, jailbreaks, and data leakage, generating a vulnerability report. Which NVIDIA tool is purpose-built for this task?
garak is NVIDIA's open-source LLM vulnerability scanner designed to automatically probe models and pipelines for weaknesses like prompt injection, jailbreaks, data leakage, and toxic output generation, producing a structured vulnerability report suitable for pre-release security review.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$129.99 $34.99 with code FREETEST33 — valid through September 7.
Get my $34.99 deal →NCP-AAI exam — quick answers
How much does the NCP-AAI exam cost?
The exam fee is approximately $200 and varies by region — confirm current pricing with the certification vendor before you book.
What topics are on the exam?
It covers 6 domains: Agent Design and Cognition (~22%), Orchestration and Multi-Agent Systems (~22%), Planning, Memory, and Retrieval (~18%), Tool Integration and Actions (~14%), Deployment, Optimization, and NVIDIA Infrastructure (~14%), Governance, Safety, and Evaluation (~10%). The full course has a dedicated chapter, lab and practice-test coverage for each.
Is this practice test really free?
Yes — all questions on this page are free with explanations and no sign-up. The paid Udemy course adds two full-length timed exams, video lessons and hands-on labs.
How do I get the discount?
Use code FREETEST33 at checkout for $34.99 (list $129.99) through September 7 — the enroll button applies it automatically.
Will this prepare me for the real exam?
The questions mirror the real exam's style and are mapped to the official domains. This is exam-focused preparation — combine the free test with the full course's timed simulations to gauge your readiness.