PMI CPMAI_v7 Exam Prep
Cognitive Project Management in AI CPMAI v7 - Training & Certification (Page 6 )

• The best choice is likely D: “Have team members that have experience, understanding of tools, and the ability to deal with massive volumes of data.”
• Why:

• Why the others are less fitting:

• For an inexperienced team and a first AI project, start with a simple, fast, and interpretable baseline.
• Naive Bayes works well for text-like data (emails) when paired with features like bag-of-words or TF-IDF, and it’s easy to train and explain.
• Establishing a solid baseline helps you understand data quality and performance before moving to more complex models.

• Neural networks are overkill for a straightforward spam classification task with limited data.
• Ensemble methods can be useful when you’re unsure which model will perform best, but they’re more complex for a first project.
• Gaussian Mixture is unsupervised and not ideal for a labeled binary classification problem like spam vs. not spam.

• Gather labeled emails and preprocess ( tokenize, clean, remove noise ).
• Convert to features (e.g., TF-IDF vectors).
• Train a Multinomial NB (or Bernoulli NB) and evaluate with metrics like accuracy, precision, recall, F1.
• Use the results as a baseline to decide if you should try more complex models later.

• The VM keeps its configuration, disks, IP settings, and associated resources.
• The operation may cause downtime.
• You cannot guarantee that it will move specifically to Host2; Azure selects the new host.
• Update management is for update configuration, not host relocation.
• Moving to another subscription does not move the VM to a different physical host.
• Usage location is unrelated to VM placement.

• Correct answer: D — AWS Migration Hub

• AWS Migration Hub provides a single location to track the progress of application migrations. It offers a centralized dashboard to view migration status across multiple tools and projects.

• AWS Application Discovery Service helps inventory and map on-prem applications and their dependencies, not a single progress tracker.
• AWS Application Migration Service (formerly SMS) performs the actual lift-and-shift migrations, but isn’t the centralized tracker itself.
• AWS Service Catalog manages catalogs of IT services for self-service deployment, not migration progress.

• Amazon Kendra is a managed enterprise search service that can index content stored in S3 and support natural-language queries to find text inside documents (PDFs, Word files, etc.).

• Rekognition analyzes images/videos, not document text search.
• Polly is a text-to-speech service.
• Lex is a chatbot service.

• Create a Kendra index.
• Add a data source of type S3 and point it to your bucket (with an IAM role that allows access).
• Configure/document conversion if needed, then index the data.
• Use the Kendra API or console to run natural-language queries against your indexed documents.

• Azure Cosmos DB is a globally distributed NoSQL database service. It stores and retrieves application data, but it does not provide the tools needed to build, test, and deploy predictive analytics models.
• Azure Machine Learning—including the historical Azure Machine Learning Studio—is designed for the machine-learning lifecycle:

• The problem: Guest users can’t create posts in the Monthly Reports channel.
• The recommended fix (answer B): Modify the Guest access settings in the Teams admin center.

• Guest access settings control what guests are allowed to do across Teams (for example, posting, chat, and channel participation). If guests are restricted from posting, enabling the appropriate guest permissions will let them post in channels like Monthly Reports.
• The other options don’t directly change guest permissions in the way that affects posting:

• BIOS language support: Useful for usability, but not whether the PC can operate electrically.
• Supply chain security: Concerns procurement and hardware trust, not deployment across countries.
• Power efficiency: Important for energy usage, but it does not ensure compatibility with local power systems.

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.

• Separate accounts/domains create a strong administrative boundary between non-production and production.
• Separate projects provide distinct IAM boundaries, billing, quotas, service accounts, and resources.
• Developers can receive permissions in the development project without automatically gaining access to production.
• Production administrators can be restricted from development resources, supporting separation of duties and least privilege.

• A: Creates excessive project fragmentation without being required by the stated requirement.
• C: Keeps development and production under one account and does not clearly separate administrative duties.
• D: Uses one account and only two projects; it provides some project isolation but weaker organizational separation.

• A. IDS attack signatures: An IDS primarily detects suspicious network activity and usually alerts rather than directly scanning and blocking uploaded files.
• C. Load balancer: Distributes traffic among servers but does not inspect files for malware.
• D. WAF: Protects web applications from attacks such as SQL injection, cross-site scripting, and some malicious requests. It is not primarily a PDF-malware scanning solution.

• Residual neural network (ResNet): Primarily used for computer vision tasks such as image classification.
• Support vector machine (SVM): Commonly used for classification and regression, not natural-language query generation.
• WaveNet: Designed mainly for generating or modeling audio and speech.

sql 
GRANT ALTER ON SCHEMA::schemaA TO User1;

• TRUNCATE TABLE is a DDL operation, not a normal DELETE.
• Having DELETE permission on a table is not sufficient to truncate it.
• The minimum relevant permission is ALTER on the table.
• Granting ALTER on SCHEMA::schemaA allows the user to truncate tables within schemaA.
• It does not grant equivalent permissions on schemaB or schemaC, satisfying least privilege.

• As the project progresses, you gain information, refine requirements, and lock down baselines (scope, schedule, cost). This reduces overall risk and uncertainty because more is known and addressed.
• At the same time, changes become more costly over time. Later changes can affect already approved baselines, cause rework, impact the critical path, require more coordination, and incur penalties or delays, all of which raise the cost of making changes.

• A assigned without var becomes a global variable (in non-strict mode). So after myFunction(), a = 5 is accessible.
• b is declared with var inside the function, so it’s local to that function (function-scoped). It’s not defined in the outer (global) scope.

• console.log(a) prints 5.
• console.log(b) throws a ReferenceError: b is not defined.

• A: Security groups do not control traffic to an S3 gateway endpoint in this way. Gateway endpoints are configured through route tables.
• B: An S3 bucket policy may be needed for authorization, but the error specifically indicates a connectivity or endpoint-routing problem.
• C: Glue connection settings do not require a fully qualified domain name to resolve this endpoint issue.

• Permanent table (E): Correct. A dropped permanent table can be restored with UNDROP TABLE during its Time Travel retention period. Permanent tables also have Fail-safe protection.
• Transient table: It can also generally be restored during its shorter Time Travel period, although it has no Fail-safe. Therefore, excluding it makes the question questionable.
• Temporary table: Usually cannot be recovered after it is dropped; its lifecycle is limited to the session.
• Materialized view (D): The answer key treats it as recoverable, but recovery behavior depends on the specific Snowflake object/version capabilities. It should not be confused with an ordinary view.

• B: Moving the task may affect the critical path and should not be done before evaluating the planned response and impacts.
• C: Excluding a required task is not appropriate without formal analysis and stakeholder approval.
• D: Lessons learned and the risk log should be updated, but only after or alongside managing the immediate issue; updating them is not the first action.

• C — “when the query finishes”

• D — “when the query began”

• B — Prompts and suggestions retained for 28 days
• C — User engagement data kept for two years

• The key distinction is between content data—your prompts, code context, and generated suggestions—and engagement data, such as usage activity.
• Do not confuse retention with training: enterprise Copilot data is generally not used to train GitHub’s models without appropriate authorization.
• For a current exam, verify the policy version cited by the exam provider. The key’s C is likely correct, but B should be treated cautiously and may no longer reflect the current policy.

powershell 
Get-AzureVM -ServiceName <service-name> -Name <vm-name> | 
  Set-AzureStaticVNetIP -IPAddress <private-ip> | 
  Update-AzureVM

• New-AzureRMVMConfig configures VM settings but does not itself assign the static IP.
• Set-AzureSubnet is not the appropriate cmdlet for assigning a VM’s private IP.
• The Azure portal option mentioned in the question was not the intended method for this older deployment model.
• Changing settings in Windows Network and Sharing Center is incorrect because Azure assigns the IP through the VM’s virtual network interface; manually configuring it inside the guest can cause conflicts.

• Voice devices ? VLAN 10
• Data devices ? VLAN 20
• SVI for VLAN 10 ? 192.168.10.1
• SVI for VLAN 20 ? 192.168.20.1

• A: Describes a physical trunk interface, not an SVI.
• B: An SVI can provide management access, but that is not its primary definition here.
• D: SVI does not compensate for insufficient physical ports; it is a logical interface.

• The user authenticates with the external Identity Provider.
• Salesforce receives the SAML assertion.
• Salesforce uses attributes in that assertion—such as username, email, first name, last name, and profile/community information—to create the user during the first login.
• On later logins, Salesforce updates or uses the existing user record.

• On-the-fly provisioning is a general phrase; the Salesforce capability being tested is specifically JIT provisioning.
• Salesforce APIs would require a separate integration and provisioning process.
• Identity Connect is designed primarily to synchronize users from an identity directory, not to create users at the moment of their first SAML login.

• Incidents & alerts — see how the phishing activity was detected, correlated, and investigated.
• Threat Explorer — search for the malicious messages, recipients, URLs, attachments, and campaign activity.
• Email and collaboration reports — assess how many users or messages were affected.
• Remediation details — verify whether messages were quarantined, removed, or otherwise contained.

• Microsoft 365 admin center — general tenant administration.
• Microsoft Entra admin center — identity and access management.
• Exchange admin center — mail-flow and Exchange configuration, rather than full phishing-incident investigation.

• Correct answer: B — From the code repository, modify Startup.cs (in ASP.NET Core you configure the Application Insights SDK in code, not with ApplicationInsights.config).

• Why the other options aren’t correct:

• How to implement fixed-rate sampling (ASP.NET Core):

var builder = WebApplication.CreateBuilder(args); 


    builder.Services.Configure<TelemetryConfiguration>(telemetryConfiguration => 
    { 
        var builder = telemetryConfiguration.DefaultTelemetrySink.TelemetryProcessorChainBuilder; 
        double fixedSamplingPercentage = 10; // 10% sampling 
        builder.UseSampling(fixedSamplingPercentage); 
    }); 


    builder.Services.AddApplicationInsightsTelemetry(new ApplicationInsightsServiceOptions 
    { 
        EnableAdaptiveSampling = false 
    }); 


    var app = builder.Build();

• Modelo de embeddings de texto: trabalha principalmente com texto, não sendo adequado para imagens.
• Modelo generativo multimodal: pode interpretar ou gerar conteúdo de vários tipos, mas não é a opção específica para busca por similaridade.
• Modelo de geração de imagens: cria imagens a partir de entradas, não realiza busca multimodal.

1. Define Power Platform DLP policies
2. Add custom prompt modifications
3. Conduct adversarial testing

1. DLP policies first

1. Prompt modifications second

1. Adversarial testing last

• Users retain their existing usernames and passwords.
• Existing accounts can be used to access Azure-based resources.
• Administrators avoid manually recreating thousands of accounts.
• The transition is less disruptive before the on-premises data center is retired.

• State is stored remotely in Terraform Cloud/Enterprise, not as a local file.
• Terraform runs (plan/apply) can be executed on remote infrastructure, with results visible in the remote workspace UI.
• Supports workspace-based organization, state locking, and concurrency control.
• Useful for team collaboration, centralized governance, and auditable logs.

• In your configuration, you use a backend block like backend "remote" { ... }.
• You still run terraform init to configure/migrate to the remote backend.
• After setup, terraform plan/terraform apply may be executed remotely, and the outputs/logs appear in the Terraform Cloud/Enterprise UI rather than (only) in your local terminal.

• Other backends store state remotely but typically do not provide remote execution; with the remote backend you get remote runs and a collaborative UI, not just remote state storage.

• '15' is a string. With + and a string, JavaScript does string concatenation: '15' + 10 -> '1510'.
• The result '1510' is then used in a multiplication: '1510' 2. The operator triggers numeric conversion, so '1510' becomes the number 1510.
• 1510 * 2 = 3020.

• If the code were let x = (15 + 10) * 2, the result would be 50 since 15 + 10 is numeric addition.

• Drops packets when traffic exceeds a certain bitrate: This describes traffic policing, not shaping.
• Fragments large packets and queues them: Shaping does not fragment packets as its primary congestion-control action.
• Drops packets randomly from lower-priority queues: Packet dropping may occur if a shaping queue becomes full, but it is not the defining behavior.

• Shaping: buffers and delays excess traffic.
• Policing: drops or marks traffic that exceeds the configured rate.

• Startup scripts run at computer startup, before any user logs on, making them ideal for machine-wide configuration on each new VM.
• They execute under the Local System account, which is appropriate for initial setup tasks.
• Linking a GPO to the OU ensures every new VM in that domain automatically receives the configuration—great for scalability.

• A) SetupComplete.cmd runs only once, right after Windows Setup, not on every boot, so it isn’t ideal for provisioning multiple VMs consistently.
• B) Logon scripts run per user, not per machine, so they aren’t reliable for initial VM configuration before a user signs in.
• D) Placing scripts in a VHD doesn’t automatically execute them.

• 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.

• prompt: the input or instruction given to the model
• completion: the desired response

json 
{"prompt":"Summarize this customer complaint.","completion":"The customer reports a delayed shipment."}

• B: A .txt file containing CSV-style lines is not automatically a valid fine-tuning dataset. The required structure and format must match the selected Bedrock model.
• C: Provisioned Throughput provides dedicated inference capacity; it does not fine-tune a model.
• D: Journals and textbooks might support domain adaptation or pretraining, but they do not provide the prompt-to-answer examples typically required for supervised fine-tuning.

1. The sender hashes the original message and sends the message plus its hash.
2. The receiver applies the same hashing algorithm to the received message.
3. The receiver compares the newly calculated hash with the sender’s hash.
4. Matching values indicate that the message was not altered in transit.

• A binary image is not the normal description of a hash result.
• Using a different algorithm produces a different type of digest, so the values cannot be directly compared.
• Hashing verifies integrity, but by itself does not prove who sent the message. Digital signatures or MACs address authentication and integrity together.

• IaaS provides virtualized infrastructure, including VMs, disks, and networking. You manage the guest operating system and software installed on the VM.
• PaaS provides a managed platform for deploying applications without managing the underlying VMs or operating systems. Examples include Azure App Service and Azure Functions.

• Context: You’ve enabled auditing in Dataverse and need to trim audit data to free storage. The question asks which deletion options you can use.

• What the options typically mean:

• About the provided answer (A, C, D): The list includes “by column,” which isn’t a typical or supported way to purge audit data. The more plausible trio is A (by record), B (date range), and C (by table older than date). If the exam expects A, C, D, that seems incorrect or mis-stated in the source.

• Practical takeaway: To reduce audit storage, use per-record deletions, per-table purge by date, and/or date-range deletions as supported by the environment’s audit management UI or admin docs.

• A. Implement retries with exponential backoff

• C. Reduce the frequency and/or size of requests

• B. Use PutRecord instead of PutRecords — PutRecord sends one record per call and does not remove the underlying throughput limitation. It may increase request overhead.
• D. Use Amazon SNS instead — SNS is not a direct fix for Kinesis shard throttling and changes the required streaming architecture.
• E. Reduce KCL consumers — KCL consumers read from Kinesis; they do not solve producer-side PutRecords throttling.

• 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.

Flat organisation

• Few levels of management
• A wide span of control—managers supervise many employees
• Shorter communication channels
• More employee autonomy and quicker decisions
• Fewer formal promotion levels

Tall organisation

• Many levels of management
• A narrow span of control—managers supervise relatively few employees
• A clear hierarchy and reporting structure
• More formal supervision and communication channels
• More obvious promotion and career stages

• Key difference: DirectQuery leaves data in the on-prem SQL Server and queries it live. That means every visual request can trigger online processing on the source, which can be slower and adds load on the on-prem system. In contrast, Import loads data into the Power BI data model, so visuals render from the in-memory engine (faster).

• Why that helps with the requirements:

• Why not A or B: Both rely on live data via DirectQuery, which contradicts the goal to minimize online processing and visuals performance.

• Why not D: It uses Import, but refreshing hourly adds unnecessary processing and complexity when daily refresh already satisfies currency needs.

• IaaS: You deploy and manage virtual machines, operating systems, disks, and networking. Azure manages the physical datacenter and host infrastructure.
• PaaS: Azure manages the operating system and runtime so you can focus on deploying applications, such as with Azure App Service or Azure Functions.
• SaaS: You consume a complete, provider-managed application.

• Windows 10 VMs
• Windows Server 2012 or later VMs
• Running VMs—they do not need to be shut down
• Debian 8.2+ VMs, assuming the specific version is supported
• Stopped or deallocated VMs

• Regional leads — Customer Community Plus

• Capacity planners — External Identity

• Customer Community does not provide the enhanced reporting/dashboard access required by regional leads.
• An Identity license is generally for internal Salesforce identity use and does not fit the external capacity-planning users as well as External Identity.

• Amazon DynamoDB — Managed NoSQL database, not PostgreSQL.
• Amazon Athena — Serverless service for querying data in S3; intended for analytics, not OLTP.
• Amazon EMR — Managed cluster platform for big-data frameworks such as Hadoop and Spark.

1. Get Data ? Folder in Power BI Desktop.
2. In Power Query Editor, expand the Attributes column to expose metadata such as hidden status and other file attributes.
3. Remove the Content column (and any other data columns you do not need).

• Name
• Extension
• Date accessed
• Date modified
• Folder Path
• Attributes
• Content

kusto 
Event 
| summarize make_set(EventID) by Computer

• summarize groups records.
• by Computer creates one result row per virtual machine.
• make_set(EventID) returns an array containing unique event IDs for that computer.

kusto 
| summarize make_list(EventID) by Computer

• Alice is co-resident on the same physical host and exploits shared hardware (the processor cache) to infer secrets. This is the classic definition of a side-channel attack—information leakage through implementation details (like timing or cache behavior) rather than by breaking a protocol or injecting code.

• Observing cache timing or access patterns to deduce cryptographic keys or plaintext data from the victim VM.

• Man-in-the-cloud attack: targets cloud trust/auth, not local hardware sharing.
• SQL injection: exploits input to run SQL commands, not a VM/caching vector.
• Service hijacking: takes control of a service, not about covert data leakage via shared hardware.

• Notice phrases like “shared hardware resources,” “co-resident,” or “cache”—these indicate a side-channel vector.