NVIDIA Agentic AI - NCP-AAI

You are creating a virtual assistant agent that needs to handle an increasingly wide range of tasks over an extended period.
What is the primary benefit of combining external storage (like RAG) with fine-tuning (embodied memory) in this context?

• A. To enhance long-term reasoning capabilities and adaptability
• B. To accelerate the agent's initial response time
• C. To ensure the agent doesn't make any errors
• D. To eliminate the need for external knowledge

A development team is building a customer support agent that interacts with users via chat. The agent must reliably fetch information from external databases, handle occasional API failures without crashing, and improve its responses by learning from user feedback over time.
Which of the following tasks is most critical when enhancing an AI agent to handle real-world interactions and improve over time?

• A. Implementing retry logic for error handling and integrating user feedback loops for iterative improvement
• B. Applying a well-structured training process with foundational generative models and prompt engineering
• C. Utilizing internal knowledge bases to support agent responses alongside external APIs
• D. Designing conversation flows that provide consistent responses based on predefined scripts

A medical diagnostics company is deploying an agentic AI system to assist radiologists in analyzing medical imaging. The system must provide AI-generated preliminary diagnoses and allow radiologists to review, modify, and approve all recommendations before patient treatment decisions. Human expertise should remain central, with detailed records of human interventions and decision rationales maintained.
Which approach would best balance human oversight with AI support in a safety-critical setting?

• A. Design a simple notification system that alerts radiologists only when AI confidence falls below predetermined thresholds, otherwise allowing autonomous operation without human review or documentation.
• B. Design a fully automated system that presents final diagnoses to radiologists for simple approval or rejection, minimizing human interaction to improve efficiency and reduce decision fatigue.
• C. Design a passive monitoring system where AI makes decisions while humans observe without ability to intervene, focusing on post-decision evaluation and quality assurance.
• D. Design an interactive system that presents AI analysis with confidence scores, allows radiologists to review evidence, modify recommendations, and requires explicit approval with documented reasoning for all decisions.

A senior AI architect at a public electricity utility is designing an AI system to automate grid operations such as outage detection, load balancing, and escalation handling. The system involves multiple intelligent agents that must operate concurrently, respond to changing data in real time, and collaborate on tasks that evolve over multiple interaction steps. The architect must choose a design pattern that supports coordination, flexible task delegation, and responsiveness without sacrificing maintainability.
Which design approach is most appropriate for this scenario?

• A. Build a rule-driven control structure that maps task flows to predefined paths for fast and efficient execution under known operating conditions.
• B. Use an agent service architecture with decoupled execution units managed by a shared interface layer that handles communication and task routing.
• C. Adopt a role-based agent model coordinated through a shared task planner, where agent decisions are informed by centralized policy logic and runtime context signals.
• D. Design the system as a stepwise sequence of agent functions, where each stage processes and passes data to the next in a fixed functional chain.

A development team is building an AI agent capable of autonomously planning and executing multi-step tasks while retaining context and learning from past interactions.
Which practice is most important to enable the agent to effectively manage long-term memory and complex tasks?

• A. Reduce planning features and memory management to keep the system streamlined.
• B. Apply short-term memory approaches that handle each interaction independently of previous ones.
• C. Use basic rule-based decision methods that emphasize fast responses over adaptive planning.
• D. Implement memory mechanisms for context retention and apply chain-of-thought prompts to enhance reasoning.