NVIDIA NCP-AAI Exam : Agentic AI

An AI engineer is evaluating an underperforming multi-agent workflow built with NVIDIA agentic frameworks.
Which analysis approach most effectively identifies optimization opportunities in agent coordination and communication patterns?

• A. Monitor workflow completion times using analysis that subsumes inter-agent communication costs, coordination overhead, and task allocation balance.
• B. Trace agent interaction patterns using observability features, measure communication overhead, identify redundant operations, and analyze task distribution efficiency.
• C. Focus exclusively on individual agent accuracy without analyzing workflow-level efficiency, coordination costs, or overall system throughput.
• D. Evaluate agents individually, allowing the toolkit to automatically infer interaction effects, communication patterns, and emergent behaviors from coordination.

You are developing a RAG solution and have decided to use a classifier branch as part of your semantic guardrail system to assess the risk of generated text.
Which of the following is a key benefit of using a classifier branch compared to solely relying on prompt filtering?

• A. Classifier branches eliminate the need for human oversight, thereby automating the safety process.
• B. Classifier branches can automatically adapt to new forms of harmful language.
• C. Classifier branches primarily focus on detecting factual inaccuracies, rather than stylistic or harmful language.
• D. Since a classifier branch does not require training, it can identify potentially problematic content.

You are rolling out a multimodal conversational agent on NVIDIA's stack: the model is containerized as a TensorRT-LLM engine, served via Triton Inference Server behind NIM microservices for routing and scaling, and protected by NeMo Guardrails for safety and compliance. During early testing, end-to-end latency exceeds your target budget, and you need to tune batching, model precision, and guardrail checks while maintaining both throughput and enforcement of safety policies.
Which configuration change is most effective for reducing latency under these constraints while still enforcing NeMo Guardrails policies?

• A. Deploy separate Triton servers for model inference and guardrail validation, routing requests sequentially and merging outputs at the application layer.
• B. Quantize the TensorRT-LLM engine to FP16, tune Triton's dynamic batching, and integrate NeMo Guardrails alongside inference to run policy checks in parallel.
• C. Quantize the TensorRT-LLM engine to INT8, disable dynamic batching, and invoke Guardrails checks synchronously within the inference path.
• D. Keep FP32 precision, increase batch size aggressively, and perform Guardrails checks in a downstream microservice after inference.

You're deploying a healthcare-focused agentic AI system that helps doctors make treatment recommendations based on patient records. The agent's reasoning is not exposed to users, and its decisions sometimes differ from clinical guidelines.
What safety and compliance mechanisms should be in place? (Choose two.)

• A. Exempt the model from compliance if it improves outcomes
• B. Allow overrides by human doctors to maintain accountability
• C. Obfuscate decision logic to protect proprietary methods
• D. Require model explainability or traceability for all outputs
• E. Prioritize autonomous speed of decision over explainability

A development team is creating an AI assistant that interacts with employees to help manage schedules and tasks. The team wants to ensure users can easily provide feedback, understand the agent's decisions, and intervene when necessary to maintain control and trust.
Which practice best supports effective human oversight and interaction with the AI agent?

• A. Incorporating user review stages before finalizing agent decisions to maintain accountability
• B. Continuously collecting and integrating user feedback throughout the agent's lifecycle to drive ongoing improvements
• C. Designing intuitive user interfaces with integrated feedback loops and transparent explanations of agent decisions
• D. Enabling flexible user interactions beyond predefined commands to accommodate diverse needs