You are tasked with deploying a multi-modal agentic system that must respond to user queries...

The selected option specifically A states “Integrate NeMo Guardrails, configure NIM microservices for optimized inference, use TensorRT-LLM for deployment, and profile the system using Triton Inference Server with multi-modal support.”, which matches the operational requirement rather than a superficial wording match. The complete stack matters: Guardrails for safety, NIM for optimized service packaging, TensorRT-LLM for inference acceleration, and Triton profiling for multimodal serving. Option A is the correct engineering choice because the requirement is not just “make the model answer,” but control the execution surface. In NVIDIA terms, TensorRT-LLM compiles optimized LLM engines; Triton schedules inference, exposes model metrics, and supports ensembles across multiple backends and modalities. The durable control mechanism is optimizing the multimodal ensemble as a pipeline, not as disconnected text, image, and audio models. That is why the other options are traps: a single model instance per GPU is rarely a complete answer because utilization depends on request shape, modality, and concurrency. For certification purposes, read the question as asking for controlled autonomy, not raw LLM creativity.