NCA-GENL Trustworthy AI Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  NVIDIA NeMo Guardrails

  Why this is correct

  NeMo Guardrails is built to add programmable safety and topical rails around an LLM at runtime, inspecting both user inputs and model outputs and applying actions such as refusing, redirecting, or rewriting responses. It works with the existing model without retraining, which matches the requirement exactly. This is the standard NVIDIA toolkit for enforcing conversational boundaries like medical, legal, and dangerous-activity refusals.

• ✗

  NVIDIA TensorRT-LLM

  Why it's wrong here

  TensorRT-LLM is an inference optimization library that compiles and accelerates LLM execution on NVIDIA GPUs, improving throughput and latency. It does not define conversational policies, inspect content for disallowed topics, or decide when to refuse a request. Using it would make the assistant faster, but the safety behavior described in the scenario must be implemented elsewhere, so it does not satisfy the requirement.

• ✗

  NVIDIA Triton Inference Server

  Why it's wrong here

  Triton Inference Server hosts and serves models across frameworks, handling batching, concurrency, and deployment concerns. It routes requests to models but has no built-in concept of topical refusal, input-output content policy, or conversational safety rails. It is an infrastructure layer, not a guardrail layer, so it cannot enforce the medical, legal, and dangerous-activity restrictions the retailer needs.

• ✗

  NVIDIA Nsight Systems

  Why it's wrong here

  Nsight Systems is a profiling tool used to analyze application and GPU performance, identifying bottlenecks in execution. It has no role in inspecting prompts or responses for policy violations and cannot block or rewrite model output. Selecting it would leave the assistant entirely unguarded, so it is unrelated to the safety requirement in the scenario.