NVIDIA NCP-AAI 問題集 : Agentic AI

In a global financial firm, an AI Architect is building a multi-agent compliance assistant using an agentic AI framework. The system must manage short-term memory for multi-turn interactions and long-term memory for persistent user and policy context. It should enable contextual recall and adaptation across sessions using NVIDIA's tool stack.
Which architectural approach best supports these requirements?

• A. Leverage NVIDIA Triton Inference Server with dynamic batching to cache session-level inputs between inference calls, and use an external Redis store for long-term memory.
• B. Leverage RAPIDS cuDF for memory tracking by streaming multi-turn conversation logs as GPU- resident data frames, assuming transactional history can be recalled and reasoned over using dataframe operations.
• C. Leverage NVIDIA NeMo Framework with modular memory management, integrating conversational state tracking, knowledge graphs, and vector store retrieval, while using LoRA-tuned models to adapt responses overtime.
• D. Rely exclusively on TensorRT to encode all prior knowledge into compiled model weights, allowing inference-only execution with no external memory dependencies across sessions.

When analyzing inconsistent performance across a fleet of customer service agents handling similar queries, which evaluation approach most effectively identifies root causes and optimization opportunities?

• A. Deploy stratified evaluation sampling across agent variants, query complexity levels, and temporal patterns while tracking decision paths using comparative analytics.
• B. Review performance across both high- and low-accuracy agent groups, comparing case outcomes and identifying patterns contributing to top and bottom results.
• C. Average performance metrics across all agents as this will smooth individual variations, query distribution differences, and temporal factors affecting agent behavior and accuracy.
• D. Assess performance data from recently improved agents and highlight strong results, using outcome comparisons to identify areas with the greatest impact on service quality.

A customer service agent sometimes fails to complete multi-step workflows when APIs respond slowly or inconsistently.
Which approach most effectively increases robustness when working with unreliable APIs?

• A. Restrict available tools to reduce decision complexity
• B. Adjust generation parameters to produce more predictable responses
• C. Cache recent API results to limit unnecessary repeated calls
• D. Add retries with exponential backoff and set request timeouts

In a production agentic system handling thousands of concurrent conversations, which state management strategy provides optimal performance while ensuring context preservation?

• A. Global shared state with locks for concurrent access
• B. Stateless design with context reconstruction from message history
• C. Session-isolated state with serialization and lazy loading

Your agent is designed to manage tasks through a service management API. The API responds with detailed event logs, but these logs contain both metadata and structured data.
To ensure the agent correctly interprets and processes the data from these logs, what's the most prudent approach?

• A. Employ a specialized parser that extracts all data fields, regardless of their type.
• B. Employing a modular design that allows the agent to dynamically adjust its parsing logic.
• C. Employ a specialized parser that adheres to the API's documentation, to insure strict adherence to structured data.
• D. Using a human-in-the-loop approach, manually inspecting and interpreting each log entry.