Question 1

When implementing stateful orchestration for agentic workflows using LangGraph, which memory management approach provides the best balance of performance and context retention?

Options :

A : Store complete conversation history in memory with periodic database syncing

B : Implement rolling window memory with fixed conversation length limits

C : Use session-ID based checkpointer with user-defined schema for selective state persistence

Answer: C

Question 2

A financial services company is deploying a multi-agent customer service system consisting of three specialized agents: a reasoning LLM for complex queries, an embedding agent for document retrieval, and a re-ranking agent for result optimization. The system experiences significant traffic variations, with peak loads during business hours (10x normal traffic) and minimal usage overnight. The company needs a deployment solution that can handle these fluctuations cost-effectively while maintaining sub-second response times during peak periods. Which NVIDIA infrastructure approach would provide the MOST cost-effective and scalable deployment solution for this variable-load multi-agent system?

Options :

A : Deploy agents directly on individual NVIDIA RTX workstations without containerization ororchestration, relying on load balancers with round-robin for traffic distribution.

B : Deploy each agent on dedicated NVIDIA DGX systems with manual scaling based on previous daystraffic predictions and static resource allocation for peak loads.

C : Deploy NVIDIA NIM microservices on Kubernetes with auto-scaling capabilities, utilizing NVIDIANIM Operator for lifecycle management and horizontal pod autoscaling based on custom metrics.

D : Deploy all agents on a single large GPU instance without containerization, scaling compute byupgrading to larger GPU instances when needed.

Answer: C

Question 3

When evaluating a multi-agent customer service system experiencing unpredictable scaling costs and performance bottlenecks during peak hours, which analysis approaches effectively identify optimization opportunities for both infrastructure efficiency and service reliability? (Choose two.)

Options :

A : Maintain consistent resource allocation across all service hours, for a more precise view of baseline traffic impact on long-term infrastructure efficiency.

B : Scale agent infrastructure based on aggregate performance trends, using system-wide monitoring toolsto identify broader optimization patterns across resources.

C : Deploy agents with configurable scaling workflows, allowing analysis of resource adjustment strategies and their effects on service stability during variable demand periods.

D : Deploy distributed tracing with cost attribution per agent type, correlating resource consumption withbusiness value metrics to identify optimization opportunities in agent deployment strategies.

E : Implement comprehensive workload profiling using NVIDIA Nsight to analyze GPU utilization patterns, identify underutilized resources, and optimize batch sizing for dynamic scaling withKubernetes HPA.

Answer: D,E

Question 4

Your team notices a spike in failed tool calls from a deployed workflow agent after a recent API schema update. The agent still returns outputs, but many are irrelevant or incomplete. Which maintenance task should be prioritized to restore accurate behavior?

Options :

A : Reset the agent’s long-term memory and reinitialize logs.

B : Update the tool function specifications and re-test action sequences.

C : Increase model temperature to encourage tool exploration.

D : Reduce tool retrieval vector similarity threshold to broaden context.

Answer: B

Question 5

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?

Options :

A : Leverage NVIDIA NeMo Framework with modular memory management, integrating conversationalstate tracking, knowledge graphs, and vector store retrieval, while using LoRA-tuned models to adaptresponses overtime.

B : Leverage RAPIDS cuDF for memory tracking by streaming multi-turn conversation logs as GPUresident data frames, assuming transactional history can be recalled and reasoned over using dataframeoperations.

C : Rely exclusively on TensorRT to encode all prior knowledge into compiled model weights, allowinginference-only execution with no external memory dependencies across sessions.

D : Leverage NVIDIA Triton Inference Server with dynamic batching to cache session-level inputsbetween inference calls, and use an external Redis store for long-term memory.

Answer: A