NVIDIA NCP-AAI Exam Prep
Agentic AI (Page 7 )

• A — conformant: Incorrect because one explicit requirement is violated.
• B — consistent: “Consistent” is not sufficient where the implementation contradicts a mandatory specification. The RPC mechanism is covered by the architecture, and the implementation differs from it.
• C — consistent: Incorrect reasoning: the RPC mechanism does have a defined architectural requirement, so conformance must be assessed.
• D — non-conformant: Correct because the implementation fails to follow the specified RPC/threading requirement.

• Product Owner: May attend, but is not required unless also working as a Developer.
• Scrum Master: May attend to help facilitate or coach, but is not required.
• Developers: Must participate because the event is specifically for them.

1. Start with the Customer table.
2. Use Merge Queries and select the Address table.
3. Match Customer[Address ID] to Address[Address ID].
4. Use a Left Outer join so every customer is retained.
5. Expand only City, State/Region, and Country.

• Append adds rows vertically; it is for stacking similarly structured tables.
• Group By summarizes rows and does not retrieve address columns.
• Transpose swaps rows and columns and is unrelated to table relationships.

yaml 
clusterIP: None

• CronJob: Runs Jobs on a schedule and normally does not need stable networking.
• Deployment: Manages interchangeable, stateless replicas and usually uses a regular Service.
• DaemonSet: Runs a Pod on selected nodes but does not inherently require stable Pod identities.

• A is correct: Days beyond the included period incur an extra daily charge.
• D is incorrect: Daily use after 10 days is not free.
• B and C: Data-transfer pricing can depend on the direction and service terms. The exam’s intended distinction is that the included device usage period is the clearly stated no-cost benefit.

• Primary and core nodes: Use On-Demand Instances because the workload is business-critical. These nodes manage the cluster and store/process HDFS data.
• Task nodes: Use Spot Instances in an instance fleet. Task nodes are suitable for interruption because they do not hold essential cluster data and can be replaced or diversified across instance types.
• Terminate the entire cluster after processing: The job runs only about six hours each morning, and EMRFS stores data in Amazon S3. Keeping the cluster running for the rest of the day would waste money.

• A: Using Spot for primary and core nodes risks interruption of the critical cluster.
• C: All On-Demand is reliable but unnecessarily expensive; terminating the cluster helps, but Spot task nodes provide greater savings.
• D: Leaving primary and core nodes running after the task finishes continues to incur costs. Savings Plans also are less appropriate for this intermittent workload.

• Honeypots: Decoy systems or services designed to attract attackers. Legitimate users should not access them, so activity on a honeypot is a strong indicator of unauthorized activity. Security teams can also observe attacker behavior.

• DNS sinkhole: Redirects requests for known malicious domains to a controlled internal server or blocks them. If an infected workstation attempts to contact command-and-control infrastructure, the sinkhole records the request and helps identify the compromised device.

• Tokenization: Protects sensitive data by replacing it with tokens.
• CI/CD: Automates software development and deployment.
• Threat modeling: Identifies potential risks before or during system design.
• Data obfuscation: Makes data harder to understand, but does not detect intruders.

1. Determine which host was infected first—the originating host.
2. Decide whether each other host is clean or infected based on communication and event logs.

• Compare timestamps and find the earliest suspicious event.
• Look for the host that initiated the malicious connection, file transfer, exploit, or command—not merely a host that later received traffic.
• Follow the timeline through the firewall:

• A host is likely infected if it both receives suspicious activity and then initiates related malicious traffic.
• A host may be clean if the firewall blocked the connection and there is no evidence of execution or follow-on communication.

• Data lakes: scalable, low-cost storage for structured, semi-structured, and unstructured data.
• Data warehouses: reliable SQL analytics, governance, and performance features.

• Data lake: Primarily provides flexible, scalable storage but may lack warehouse-style reliability and governance.
• Data warehouse: Strong for structured analytics but less flexible for raw and unstructured data.
• All of these: The individual lake and warehouse do not, by themselves, unify the siloed architecture.

• A — Reports: Reports are informational and don’t enforce access during an SSO transaction.
• B — Apex trigger and Permission Sets: A trigger could update permissions when a Case changes, but this creates synchronization and timing issues. It also changes Salesforce permissions rather than making a direct, real-time decision for the external system.
• C — Custom SAML JIT Provisioning: JIT provisioning is primarily for creating or updating users at login. It isn’t the appropriate mechanism for dynamically querying business records and deciding whether access should be granted.
• D — Connected App Handler: Designed for custom, programmatic control during connected-app authentication and authorization. Apex can check the user’s current open classified cases and permit or reject the SSO request.

• s3:GetObject for the relevant objects
• s3:ListBucket if listing is required
• Appropriate bucket and object resource permissions

• B: Making the bucket public is insecure and unnecessary.
• C: Prompt engineering cannot override AWS IAM or encryption permissions.
• D: Data sensitivity does not determine whether the role can access it.

1. Identify the fact table

• Records something that happened, such as a sale, test, appointment, or measurement
• Contains foreign keys to other tables
• Contains measures or numeric results
• Has many rows

2. Identify a dimension table

• Describe people, products, dates, locations, or departments
• Contain descriptive columns
• Are referenced by the fact table
• Have fewer rows than the fact table

3. Identify the surrogate key

• System-generated
• Unique within the dimension
• Often named PatientKey, CustomerKey, or similar
• Usually an integer
• Not meaningful outside the model

• Prevent connections to malware command-and-control servers
• Identify infected systems making those DNS requests
• Collect information for incident response

• A: Attackers do not normally use sinkholes to map a company’s domain structure.
• B: Redirecting employees from legitimate sites to attacker-owned sites describes malicious DNS manipulation, not a defensive sinkhole.
• D: Attracting attackers away from real resources is more closely associated with a honeypot or deception system.

• Use private connectivity from a cloud provider
• Set up SSO for federated authentication

• Private connectivity—such as a cloud provider’s private endpoint—keeps Snowflake access within the company’s private network path instead of exposing access over the public internet. This fits the VPN/VDI requirement.
• SSO for federated authentication allows Snowflake users to authenticate using the credentials already managed by the company’s identity provider—the same identity system used for VDI—eliminating separate Snowflake passwords.

• Tri-Secret Secure protects encryption keys; it does not provide private network access or credential reuse.
• A proxy Snowflake account outside the VPN conflicts with the requirement that Snowflake have no internet visibility and is not the normal solution for federated login reuse.

• Create and manage approved IT services as products
• Control which users can deploy those products
• Apply governance, permissions, and constraints
• Use and manage infrastructure-as-code templates, such as AWS CloudFormation templates

• AWS Resource Explorer: Helps find and browse AWS resources; it does not govern IaC deployments.
• AWS Organizations: Manages accounts, consolidated billing, and organization-wide policies.
• AWS Systems Manager: Provides operational management, such as patching, automation, and parameter storage, but it is not primarily used to catalog and govern IaC-based service deployments.

• Manage multiple organizations from one hierarchy.
• Enterprise-wide policies, including repository visibility, forking, and GitHub Copilot policies.
• Centralized identity and access management, such as SAML SSO, SCIM provisioning, and—where applicable—Enterprise Managed Users.
• Centralized billing and licensing, including organization and Copilot seat management.
• Enterprise audit log, which aggregates activity across organizations.
• Security and compliance oversight, such as organization security visibility and enterprise-wide controls.
• Enterprise support and account management features, depending on the plan.

• Organization: manages its own repositories, teams, members, and local policies.
• Enterprise: governs multiple organizations and can enforce policies across them.

• Create Power Query parameters for the Azure SQL server/database—or at least the database name.
• Use the parameter in the PBIX connection.
• Publish the same PBIX separately to the relevant development, test, and production workspaces.
• Set each dataset’s parameter value in the Power BI service so it points to the corresponding Azure SQL database.

• Development dataset ? DevelopmentDB
• Test dataset ? TestDB
• Production dataset ? ProductionDB

• The same PBIX is used for every lifecycle stage.
• Each stage consumes data from its matching database.
• Deployment pipelines are not required.
• Azure SQL is a cloud source, so gateway administration is generally avoided.
• Only the parameter value differs between published datasets, reducing maintenance.

• A — Correct: Flow Builder is declarative and visual, so administrators and other non-developers can participate in designing and improving business processes without writing Apex.
• D — Correct: Flow Builder provides visual debugging tools, allowing an admin to test paths, inspect variable values, and identify where a specific business case fails.
• E — Correct: A flow’s visual structure documents the logic and makes later maintenance easier than reading and updating custom Apex code.

• B — Incorrect: Flow debugging does not attach a debugger to live customer sessions. The admin uses Flow Builder’s debug functionality with test inputs; this is not live-session debugging and should not be used to monitor real customer transactions.

• Detect risky sign-ins from unfamiliar or unidentified IP addresses
• Evaluate user or sign-in risk
• Automatically require a password change because of a risky sign-in

• User risk policy: Can require a user to securely change their password when the account may be compromised.
• Sign-in risk policy: Can require MFA or block access based on the risk of a particular sign-in.

• Input feature distributions differ from training data ? Analyze dataset drift metrics

• Model accuracy drops without code changes ? Analyze dataset drift metrics

• Endpoint latency increases under load ? Investigate scaling and infrastructure metrics

• B — AWS DataSync: Transfers data between storage systems, such as S3, file systems, and on-premises storage. It is not a marketplace or catalog for third-party datasets.
• C — Kinesis Data Streams and CodeCommit: CodeCommit is a source-code repository, not a third-party data provider. Kinesis is for streaming data ingestion.
• D — Kinesis Data Streams and ECR: ECR stores container images, not datasets. Kinesis does not provide access to ECR data as an external dataset source.

• Exploitability
• Attack complexity
• Required privileges
• User interaction
• Impact on confidentiality, integrity, and availability

• CVE: Identifies and tracks a specific vulnerability; it does not primarily measure severity.
• OSINT: Open-source intelligence gathered from publicly available sources.
• SOAR: Security orchestration, automation, and response technology used to automate security workflows.

• Resources that do not need internet access should remain on private network segments.
• Internet-facing access should be restricted to systems that require it.
• Access should be explicitly allowed rather than broadly exposed.

• A: A single administrator with access to core functions may violate least privilege if that access is broader than necessary.
• B: Inspecting traffic supports monitoring and defense, but does not itself limit privileges.
• C: Updating routing configurations supports availability and connectivity, not least privilege.

• Authentication — verifies who the user is.
• Authorization — determines what the authenticated user is allowed to access.
• Accounting — records session activity, such as login time and resource usage.

• CIA — confidentiality, integrity, and availability; core security objectives.
• ACL — access control list; rules that permit or deny traffic or access.
• PEM — Privacy Enhanced Mail, an older email security standard.

• Microsoft 365 groups can contain users, but generally can’t contain other groups as members.
• Security groups can contain users and, where supported, other security groups.
• Distribution groups have different nesting restrictions and aren’t equivalent to security groups.
• Dynamic groups can’t be manually populated with members; their membership is calculated from rules.
• A group’s membership type (Assigned or Dynamic) and group type determine whether it can be added to another group.

• Virtual machine disks
• Databases
• Storage accounts
• File systems

• Site-to-site VPN: protects data moving between networks
• HTTPS: protects data moving between a client and website
• Encrypted email: generally protects the message while being transmitted

• At rest = stored data
• In transit = moving data
• In use = data currently being processed by a system

• Normalize casing in the data source—for example, convert all values to uppercase or lowercase.
• Add an index key to provide a unique, stable identifier for each row or value.

• B — Increase provisioned concurrency for Lambda

• D — Refactor the Lambda functions

• E — Use Amazon RDS Proxy

• The listener has a virtual IP address on the load balancer.
• The active SQL Server replica must respond using the listener’s IP address, not its own VM IP.
• Floating IP allows the listener IP to be preserved and returned by the active replica.
• This is required for the SQL Always On listener configuration.

• A health probe for the SQL listener/cluster role
• The SQL Server listener port, commonly 1433
• Appropriate backend pool membership

• 25 — Traditional SMTP server-to-server mail transfer; not normally used for secure client submission.
• 110 — POP3, used to retrieve email and typically unencrypted.
• 143 — IMAP, used to access email and typically unencrypted.

• SAML: Enables federated authentication and single sign-on between the IdP and SaaS providers.
• OAuth tokens: Allow delegated application access; OAuth is not primarily the first step for employee SSO.
• Password vaulting: Stores or autofills multiple passwords but does not eliminate the separate credentials or provide true federation.

Why

cmd 
PowerShell.exe -ExecutionPolicy Bypass -File C:\Scripts\ConfigureVM.ps1

Why the other options are less suitable

• GPO logon script: Runs only when a user logs on and may not run during initial VM provisioning.
• GPO startup script: Can work, but requires the VM to be joined to the domain and the policy to be applied. It is also normally evaluated at each startup.
• Place scripts in a VHD: Merely placing scripts on a disk does not cause them to execute.

• An Azure pay-as-you-go subscription follows an operational expenditure (OpEx) model.
• With OpEx, the organization pays for services as they are used, typically based on consumption.
• There is no need to purchase and maintain physical servers upfront.
• This contrasts with capital expenditure (CapEx), where the company makes a significant upfront investment in infrastructure, such as on-premises Hyper-V hosts.

1. Ingestion Lambda receives requests from API Gateway and places messages in an SQS queue.
2. Processing Lambda is triggered by the queue and loads records into Aurora PostgreSQL.

• Decoupling: API request handling is separated from database loading.
• Buffering: SQS absorbs bursts when incoming traffic exceeds the database’s processing capacity.
• Controlled concurrency: Configure the SQS-triggered Lambda’s concurrency and batch size to avoid overwhelming Aurora or exhausting database connections.
• Automatic retries: Messages that fail processing can return to the queue and be retried. A dead-letter queue can capture persistent failures.
• Lower operational effort: The solution remains serverless and avoids managing EC2 servers or Kubernetes clusters.

• A: EC2 and Tomcat add server-management overhead.
• B: DynamoDB/DAX would require changing the database model and does not solve the Aurora ingestion design.
• C: SNS is useful for notifications and fanout, but SQS is better for durable buffering and backpressure in this producer-consumer workflow.

• Valor maior: mais aderência às instruções e aos detalhes descritos.
• Valor menor: mais liberdade criativa e resultados potencialmente mais variados.
• Valor excessivamente alto: pode causar artefatos ou imagens menos naturais.

• Aumentar o número de passos de geração: pode melhorar qualidade e refinamento, mas não é o principal controle de especificidade do prompt.
• MASK_IMAGE_BLACK: relaciona-se ao uso de máscara em geração ou edição de imagens, não à especificidade textual.
• Prompt strength: é mais associado à força do prompt em processos como image-to-image; nesta questão, o parâmetro diretamente ligado à aderência textual é o CFG scale.

• SOW
• SLA
• MOA
• MOU

• Import — Loads data into Power BI. It requires a refresh, so it may not reflect changes within two minutes unless refresh is configured appropriately.
• Live connection — Connects to an existing Power BI semantic model or Analysis Services model; it is not the normal choice for connecting directly to an Azure SQL database.

• Identity provider (IdP): Authenticates users and asserts their identity.
• Service provider (SP): The application being accessed—in this case, Salesforce/the employee portal.
• Identity store: A directory that stores user identities, such as Active Directory.
• Authentication store: A system used to validate credentials.

• Leaf switches provide access ports for servers, endpoints, and other devices.
• Spine switches form the fabric’s backbone.
• Every leaf connects to every spine, creating a predictable, high-bandwidth, low-latency path.

• A: Adding both a spine and leaf is unnecessary for simply adding access ports.
• B: Adding a spine does not directly provide endpoint access ports.
• D: A single connection to one spine creates a bottleneck and a single point of failure.

• 800 devices ? a scalable, centralized solution is needed.
• Four hardware/vendor classes ? the tool must handle different drivers and images.
• Two user groups ? it should allow different configurations for each group.
• “Distribute custom images” ? an imaging/deployment tool is required, not just a file-copy utility.

• A. Microsoft 365 group with Assigned membership
• C. Microsoft 365 group with Dynamic User membership

Why?

• Assigned membership: You explicitly add User1, User2, and User3.
• Dynamic User membership: Membership is automatically determined by a rule based on user attributes.

Why the other options are incorrect

• B and D: Security groups do not support the Microsoft 365 group expiration policy.
• E: Dynamic Device groups are intended for devices, not these users, and are also security groups.

• The VPN gateway is in VirtualNetworkA.
• Gateway transit is configured on the peering associated with the VNet that contains the gateway—here, VirtualNetworkA.
• A peered VNet may use the remote gateway through the appropriate peering option, but this does not automatically update an existing P2S client’s routes.
• The Windows 10 workstation’s P2S VPN client package was created before VirtualNetworkB was peered, so it may not have a route for VirtualNetworkB.

• Replace the keyboard: More expensive and unnecessary before simpler causes are ruled out.
• Replace the key switch: Laptop keyboards generally aren’t designed for individual switch replacement.
• Replace the circuit board: Excessive for a single nonworking key and not a cost-effective first step.

• Which Security policy matches
• Which NAT rule/translation is selected
• Which PBF rule applies
• Which route is used

text 
test security-policy-match ... 
test nat-policy-match ... 
test pbf-policy-match ... 
test routing fib-lookup ...

• View/inspect configuration: Possible.
• Modify the installed application or its S-Controls: Typically restricted in a test-drive org.

• Personally identifiable information (PII) detection

• Prompt Shields

• Self-harm content filtering addresses self-harm-related content.
• Violence content filtering addresses violent content.
• Neither protects customer PII or detects document-based prompt injection.

• GPU: A failure may cause no video output, but the PC usually still powers on.
• RAM: Can prevent booting or produce error beeps, but fans and power indicators generally work.
• CPU: Can prevent startup or booting, but it usually does not cause a completely lifeless system.

• Customer contains one row per customer and the customer’s Address ID.
• Address contains one row per address and the corresponding location details.

1. Select the Customer query.
2. Choose Merge Queries.
3. Match Customer[Address ID] with Address[Address ID].
4. Use a Left outer join so every customer is retained.
5. Expand the merged Address column and select City, State/Region, and Country.

• Append adds rows vertically and is used for tables with the same structure.
• Group By summarizes rows rather than adding related columns.
• Transpose switches rows and columns and does not establish a relationship.

1. Convert text and images into numerical embeddings.
2. Store or compare those embeddings.
3. Retrieve the most semantically similar items.

• Embedding model: Best for similarity search and retrieval.
• Generation model: Best for producing new text or other content from multimodal input.