AI-102 Search and Knowledge Mining

To define a knowledge store in Azure Cognitive Search that includes both the social media posts and the Sentiment Analysis results, you should include the following two fields in the definition:

A. storageContainer: This field specifies the Azure Blob Storage container where the enriched data will be stored. You need to define a storage container to hold the processed data from the enrichment pipeline.

B. storageConnectionString: This field provides the connection string to the Azure Storage account that will be used to store the enriched data. The connection string is necessary for Azure Cognitive Search to connect and store the data in the specified storage account.

Both the storageContainer and storageConnectionString fields are required to set up a knowledge store in Azure Cognitive Search that can store the social media posts and the Sentiment Analysis results.

This is about skillset configuration - not knowledge store config.


"name": "my-skillset",
"skills": [ ... ],
"knowledgeStore": {
"storageConnectionString": "...",
"projections": [
{
"objects": [ ... ]


you must have a connectionstring (to the destination)
you must define the type of projection: in this case object (media posts and sen.analysis are text)

ChatGPT 5.2:

"The confusion here comes from mixing what is mandatory with what is optional.
A knowledge store always requires storageConnectionString. Without it, nothing can be persisted.
For sentiment analysis scenarios, Microsoft’s official labs and documentation project enriched content into tables (structured scores, labels, analytics).
storageContainer, objects, and files are only required for specific projection types and are optional, not mandatory, for defining a knowledge store.
Therefore, the exam-safe answer is B (storageConnectionString) and D (tables)."

The two correct fields are: B. storageConnectionString - Required to connect to Azure Storage account D. tables - Optimal projection type for storing social media posts and sentiment analysis results

These two fields work together to define a knowledge store that can persist the enriched social media data and sentiment analysis results in a structured, queryable format within Azure Table Storage.

B. storageConnectionString: When defining a knowledge store, you must provide a connection string for Azure Storage. This specifies where the data will be stored.
[https://docs.azure.cn/en-us/search/knowledge-store-concept-intro?tabs=portal](https://docs.azure.cn/en-us/search/knowledge-store-concept-intro?tabs=portal)

D. tables: When storing social media posts and Sentiment Analysis results, the "tables" projection is typically used. You can store each document and its analysis results (such as sentiment scores) in a table.
[https://microsoftlearning.github.io/AI-102-AIEngineer/Instructions/24-knowledge-store.html](https://microsoftlearning.github.io/AI-102-AIEngineer/Instructions/24-knowledge-store.html)

1/ To create the KB, you just need the storageConnectionString
See https://learn.microsoft.com/fr-fr/azure/search/knowledge-store-create-rest
2/ You have then to define the right projection. In this case, I'm not totaly sure, but I would say "objects".
extract of https://learn.microsoft.com/fr-fr/azure/search/knowledge-store-projections-examples :

"knowledgeStore" : {
"storageConnectionString": "DefaultEndpointsProtocol=https;AccountName=<Acct Name>;AccountKey=<Acct Key>;",
"projections": [
{
"tables": [ ],
"objects": [ ],
"files": [ ]
}
]
}

tables (for storing the sentiment analysis results)
objects (for storing the posts in JSON format)
In an Azure Cognitive Search knowledge store, enriched data from your pipeline can be projected into three types of storage:

Tables → for structured data (like sentiment analysis results).

Objects → for JSON documents (like the original social media posts).

Files → for images or documents, but not needed here.

Correct answers:
A. storageContainer
Specifies the Azure Blob Storage container where enriched output will be stored.

D. tables
Used to define structured outputs (like sentiment analysis results) in table format for downstream querying.

Why not the others?
B. storageConnectionString
This is part of the cognitive search skillset or pipeline configuration, not the knowledge store definition itself.

C. files
Used for raw file output like images or PDFs, not ideal for structured data like social posts and sentiment analysis.

E. objects
Used when outputting JSON objects, but not specifically required for the typical structure of a sentiment analysis pipeline

There is another question where the user case is to store social media posts in tables.

To define a knowledge store that includes both the social media posts and the Sentiment Analysis results, you should include the following two fields:

B. storageConnectionString D. tables

These fields will allow you to store the analyzed data in a structured format and connect to the storage service.

BE is the answer

To define a knowledge store that includes social media posts and Sentiment Analysis results, you should include the following fields:
-tables: This field is used to store structured data, such as the results of the Sentiment Analysis.
-objects: This field is used to store unstructured data, such as the social media posts themselves.

StorageConnectionString is not a knowledge store. It can't be B.

Selected Answer: BE

was on exam 20.06.2024.

I say this answer is B and E.

It is obvious that BE is the correct answer.

AB: We need storage to upload data and connection string. https://learn.microsoft.com/en-us/azure/search/knowledge-store-create-rest

Answer is correct however... the Entity Recognition skill is now discontinued replaced by Microsoft.Skills.Text.V3.EntityRecognitionSkill.

Also categories will only allow "Person" as valid category type - not "Persons".

This (old) version allows output to be either "entities" or "namedEntities"
New version only allows "namedEntities" from answer list.

https://docs.microsoft.com/en-us/azure/search/cognitive-search-skill-entity-reco

I also got this on July 3 2023 exam, but it's changed to Skills.Text.V3.

Second answer should be namedEntities -- there is no "entities" for outputs: https://learn.microsoft.com/en-us/azure/search/cognitive-search-skill-entity-recognition-v3

person and namedEntity

This question was in the test I took in August 2024

Is this question still available on May 21, 2024?

Is this question still available on May 21, 2024?

The document processing solution aims to extract information related to persons, locations, and organizations.
Additionally, it should recognize named entities and assign them to appropriate categories.

A modified version of this was in the exam today.

Agreed

The first answer should be no:
https://learn.microsoft.com/en-us/azure/search/cognitive-search-output-field-mapping?tabs=rest
"Output field mappings are required if your indexer has an attached skillset that creates new information, such as text translation or key phrase extraction. During indexer execution, AI-generated information exists in memory only. To persist this information in a search index, you'll need to tell the indexer where to send the data."

According to my source, the code needs 'outputFieldMappings ', and that is not the case in this example.

No
Yes
No

YYN. From https://learn.microsoft.com/en-us/azure/search/cognitive-search-concept-intro: " Index content is populated from skill outputs,"

1. "CompanyDescription is available for indexing."
- "Yes". In the schema, "companyDescription" is defined as an output of the custom skill. Outputs from skills are available for indexing or for use by other skills in the skillset.

2. "The definition calls a web API as part of the enrichment process."
- "Yes". The "@odata.type" is set to "#Microsoft.Skills.Custom.WebApiSkill" and there's a "uri" field pointing to "https://contoso-webskill.azurewebsites.net/api/process". This indicates the skill calls an external web API to perform its enrichment.

3. "The enrichment step is called only for the first organization under '/document/organizations'."
- "No". The "context" field is set to "/document/organizations/*" where the asterisk (*) wildcard indicates the skill will be called for each organization under the path, not just the first one. The asterisk is a path wildcard that processes all items in a collection.

I've never seen indexing by a description , i guess you index by a unique data as id or email , a description can be duplicated in 2 or more rows so NYN

YYN
output field must be mapped in index, otherwise how will it use in its output to user.. so yes it will be index so that users can query the field

N, Y, N

CompanyDescription is available for indexing: Yes. The companyDescription is defined in the outputs section of the custom skill, which means it will be produced by the skill and can be used in subsequent steps of the enrichment pipeline, including indexing.
The definition calls a web API as part of the enrichment process: Yes. The uri field in the custom skill definition is set to https://contoso-webskill.azurewebsites.net/api/process, which is a web API endpoint. This means that the custom skill will call this web API as part of the enrichment process.
The enrichment step is called only for the first organization under “/document/organizations”: No. The context and source fields are set to /document/organizations/*, which means the custom skill will be applied to each organization under /document/organizations, not just the first one.

Final Answer:
NYN

I think it’s No for the first question. It’s basically testing you if you know this: There are no predefined outputs for this skill. Be sure to define an output field mapping in the indexer if the skill's output should be sent to a field in the search index. So by default, it is not available for indexing. It can be available if you do whatever needs to be done as the above described.

N, Y, N
https://learn.microsoft.com/en-us/azure/search/cognitive-search-defining-skillset#send-output-to-a-destination
Although skill output can be optionally cached for reuse purposes, it's usually temporary and exists only while skill execution is in progress.
• To send output to a field in a search index, create an output field mapping in an indexer.
• To send output to a knowledge store, create a projection.

https://learn.microsoft.com/en-us/azure/search/cognitive-search-output-field-mapping?tabs=rest
In contrast with a fieldMappings definition that maps a path between two physical data structures, an outputFieldMappings definition maps in-memory data to fields in a search index.

The first answer is NO. This code snippet is a skill definition. The outputs need to be mapped before they are available for indexing:
https://learn.microsoft.com/en-us/azure/search/cognitive-search-concept-annotations-syntax

I'd say the box1 is NO. The property is available for mapping to an index property.

Question says : ..."You need to configure an enrichment pipeline to perform optical character recognition (OCR) and "text analytics"... Just because of this second requirement Answer is a Multi Cognitive Service (D), alone vision service (=OCR wont make any text analytics, text extraction is not a text analytics !).. Read here : https://learn.microsoft.com/en-us/azure/search/cognitive-search-attach-cognitive-services?tabs=portal

For processing 50,000 documents with OCR and text analytics in an enrichment pipeline, while still aiming to minimize costs, a more practical approach would be:

D. a new Cognitive Services resource that uses the S0 pricing tier
This option is more suitable for several reasons:

Capacity and Scalability: The S0 pricing tier offers a higher transaction limit, making it more suitable for processing a large volume of documents like 50,000. It can handle the scale of operations required for this task without hitting the cap as quickly as the free tier might.

Cost Efficiency at Scale: While the S0 tier incurs costs, it is designed to be cost-effective for medium to large-scale operations. By carefully managing the use of resources and optimizing the processing pipeline, it's possible to keep costs under control while meeting the project's requirements.

The correct answer is:

## ✅ **B. a free (Limited enrichments) Cognitive Services resource**

### **Why this is the correct choice**

* Azure Cognitive Search allows **20 free built‑in enrichments per indexer per day** when you attach a **free Cognitive Services resource**.
* OCR and Text Analytics in a skillset both use **built‑in skills**, backed by Cognitive Services.
* Since the requirement states **“minimize costs”** and you have **50,000 scanned documents**, you want to avoid provisioning paid resources unless necessary.
* Attaching the **free (limited enrichments)** Cognitive Services resource lets you run the enrichment pipeline without incurring S0 tier costs.

B. a free (Limited enrichments) Cognitive Services resource

This is the correct answer to minimize costs. Here's why:

Free Cognitive Services resource provides:

Up to 20 transactions per indexer per day for built-in skills
Includes OCR and text analytics capabilities
No additional cost beyond the Azure Cognitive Search service itself

Limited enrichments, which are sufficient for small to medium-scale workloads or testing scenarios.

it's mentioned about minimize costs, why B is not considered as a choice, because it have 20k a day for free, 20 + 20 + 10?

correct answer should be A beacuse Computer Vision is the Azure service that includes OCR capabilities (Optical Character Recognition). When you're dealing with scanned documents and need to perform OCR as part of your enrichment pipeline, you'll typically use Computer Vision.

I checked with many LLM services and get answer B:
a free (Limited enrichments) Cognitive Services resource

Copilot says

A. a new Computer Vision resource

Explanation:

To perform optical character recognition (OCR) on scanned documents, you would need to use the Computer Vision service, which is part of Azure Cognitive Services. This service can extract printed and handwritten text from images and documents. It's also capable of extracting rich information from images to categorize and process visual data.

The Text Analytics service, also part of Azure Cognitive Services, can be used for processing the text once it's extracted, but the question specifically asks about the OCR part of the pipeline, so the Computer Vision service is the correct choice.

The cost would depend on the volume of data processed, but using a dedicated Computer Vision resource is likely to be more cost-effective than using a general Cognitive Services resource or an Azure Machine Learning Designer pipeline. The free tier of Cognitive Services is not suitable for processing a large volume of documents like 50,000.

A Multiservice account is required since both OCR and text analytics are required.
Since a multiservice account does not support the free tier (F0), then B is not a valid option. So, the answer has to be D.

Use S0 SKU

D is right.

Given the goal of minimizing costs while performing OCR and text analytics on a collection of 50,000 scanned documents, and considering that the Limited enrichments tier offers basic enrichments at no additional cost, the most suitable choice is indeed:

B. a free (Limited enrichments) Cognitive Services resource.

B. a free (Limited enrichments) Cognitive Services resource

Explanation:

The free (Limited enrichments) tier of Cognitive Services provides access to Optical Character Recognition (OCR) and basic text analytics capabilities.
Attaching this Cognitive Services resource to the skillset allows you to perform OCR and text analytics on the scanned documents.
Since the requirement is to minimize costs, choosing the free tier is the most cost-effective option.
This option aligns with the goal of making the text available through Azure Cognitive Search while keeping costs low.

Correct option is D - a new Cognitive Services resource that uses the S0 pricing tier

We can't do this with free tier, thus we can't select Option B

to minimize cost you would not use a S0 tier which is why the answer is A

OCR is a built-in skillset that required billable tier of Azure AI Search, so the D is correct

Using a higher tier does not imply scaling replicas or partitions.
The answer should be no

answer is correct

Azure Cognitive Search throttling commonly occurs when you exceed the capacity limits of your current service tier (such as queries per second, partitions, or replicas).

Migrating to a higher service tier:

Increases available query throughput

Allows more replicas and/or partitions
Raises service limits that directly impact throttling

a is correct

Shouldn't you increase replicas first?

I'd say NO because the question does not tell us that we have reached the replica limit in the existing tier

Yes, migrating to a higher tier in Azure AI Search can help reduce throttling, but it’s often more cost-effective to scale out by increasing the number of replicas instead. Adding replicas improves query concurrency without the higher expense associated with moving to a higher tier, making it a generally more economical approach. It's true anyway.

"Basic provides dedicated computing resources for production workloads at a smaller scale, but shares some networking infrastructure with other tenants." -> changing tier allows to use dedicated networking infrastructure, so it reduces the throttling. https://learn.microsoft.com/en-us/azure/search/search-limits-quotas-capacity

Yes,
Migrating to a Cognitive Search service with a higher tier can reduce the likelihood of search query requests being throttled because higher tiers generally offer more resources and higher request limits.

not add replicas?

Right answer is No.

This is correct to use a replica, so the answer is no. All but M7, T20, and T25 graduates must give a silly answer of yes.

I would say NO, the reason is that a higher tier does not always increase the resources, it depends from which tier you are incresasing. However, this question later comes with the option of increase the replicas:
A simple fix to most throttling issues is to throw more resources at the search service (typically replicas for query-based throttling, or partitions for indexing-based throttling). However, increasing replicas or partitions adds cost, which is why it's important to know the reason why throttling is occurring at all. Investigating the conditions that cause throttling will be explained in the next several sections.

The best answer is:
• A. Yes
Reason for choosing this option: Azure Cognitive Search service tiers determine the resources available for that service. Higher tiers provide more resources such as increased document counts, indexing speed, and query capacity. If the query volume has increased and requests are being throttled, it indicates that the current tier’s resources are insufficient to handle the load. Migrating to a higher tier will provide more resources, reducing the likelihood of throttling.

throttle is a simple performance tier issue, using higher tier will resolve the issue

Choose higher tier, see https://learn.microsoft.com/en-us/azure/search/search-sku-tier

A. Yes.

Migrating to a higher tier in Azure Cognitive Search can provide more resources, such as increased storage, throughput, and replicas, which can help reduce the likelihood of search query requests being throttled.

Answer is no. you should create a private link with private endpoint

- "You create a web app named app1 that runs on an Azure virtual machine named vm1. Vm1 is on an Azure virtual network named **vnet1**."
-"You deploy service1 and a public endpoint to a **new virtual network**, and you configure Azure Private Link."
It seems that app1 and service1 are on two different virtual networks. Since it doesn't mention that they are paired, traffic between them goes through the Internet, even if service1 has a private endpoint in the new vnet.
https://learn.microsoft.com/en-us/azure/virtual-network/virtual-networks-udr-overview

Answer is Yes

private endpoint should be used

Yes, this solution meets the goal. Even though service1 technically has a public endpoint, using Azure Private Link allows your app to access it privately, through your own virtual network (vnet1).

Private Link creates a private IP address for service1 inside your network.

So, app1 can talk to service1 entirely within the Azure private network, just like talking to another server in the same room — no need to go out onto the public internet. Private Link lets you access Azure services securely and privately from inside your virtual network.

Even though service1 has a public endpoint, Private Link overrides that with a private connection.

So traffic stays inside Azure’s private backbone, not the internet.

The solution specifies deploying service1 with a public endpoint. To ensure app1 can connect directly to service1 without routing traffic over the public internet, you should deploy service1 with a private endpoint instead. The private endpoint will allow secure connectivity through Azure Private Link, enabling app1 on vnet1 to communicate directly with service1 over the private network.

Here’s the correct approach:
- Deploy service1 to a virtual network (preferably the same vnet1 or a peered network).
- Use a private endpoint for service1.
- Configure Azure Private Link to enable secure, direct access.

This configuration ensures that the connection does not use the public internet.

Answer is A,YES. This core of any cloud network facing public internet to private interface. Solution says: "deploy service1 and a public endpoint to a new virtual network, and you configure Azure Private Link"---Virtual networks are configured with private IP/endpoint which will then be recognizable under Azure Private Link

Yes is the correct answer.
By deploying service1 and configuring a public endpoint within a new virtual network, and then enabling Azure Private Link, you ensure that the connection between app1 (on vm1) and the search service (service1) happens over a private network.

This is so stupid and confusing. If you deploy the service first with public endpoint and after you configure a private link endpoint for it, it will automaticly close don the public endpoint. So yes its does meet the goal. But these questions are there to trick you.

This answer is No.

This is correct answer is no. All but M7, T20, and T25 graduates must give a silly answer of yes.

Answer is a big YES. Like other PaaS services in Azure, you can deploy your service with public access (i.e. in this case they just call it public endpoint) and configure your private access after with a private endpoint. Doing so, the private endpoint will block the public access.

Private endpoints for your search service enable you to:

Block all connections on the public endpoint for your search service.

https://learn.microsoft.com/en-us/azure/search/service-create-private-endpoint#why-use-a-private-endpoint

A is correct!
That solution would work perfectly. Here's why: By deploying the search service and the endpoint to the same virtual network, you create a private, secure connection between the web app and the search service. And by using Azure Private Link, you're creating a private, dedicated connection between the two services, which means traffic doesn't have to go over the public internet. This not only increases security but also improves performance and reliability. So, deploying the service to a new virtual network and configuring Azure Private Link would definitely meet the goal of connecting app1 directly to service1 without using the public internet.

Not a public endpoint but a private endpoint.

"You can use private endpoints for your Azure AI services resources to allow clients on a virtual network to securely access data over Azure Private Link. "
https://learn.microsoft.com/en-us/azure/ai-services/cognitive-services-virtual-networks?tabs=portal#use-private-endpoints

appeared in august exam

B is correct answer : No.

Public endpoint means the traffic is going to Internet, so answer is a big 'NO'

Y
N
Y

https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/entities-linking/entities-linking?tabs=HTTP

The code will detect the language of documents.
No. The provided code doesn't include any language detection functionality. Azure Text Analytics does have separate language detection capabilities, but they aren't used in this snippet.
The url attribute returned for each linked entity will be a Bing search link.
No. The RecognizeLinkedEntities function in Azure Text Analytics usually returns links to more authoritative sources like Wikipedia, not general Bing search results.
The matches attribute returned for each linked entity will provide the location in a document where the entity is referenced.
Yes. The matches attribute within the response from RecognizeLinkedEntities provides details about where each recognized entity occurs in the input text. This includes information like character offsets, allowing you to locate the entity within the original text.

YNY
1) The API returns the language for each found document
https://learn.microsoft.com/en-us/azure/ai-services/language-service/entity-linking/quickstart?tabs=windows&pivots=rest-api#json-response
2) As per samples in Learn the url is always the link to Wikipedia
3) Matches contains information about the start and end indexes of each match

N N Y

LinkedEntity.Url Property - Gets the URL that identifies the linked entity's entry in the data source.
https://learn.microsoft.com/en-us/dotnet/api/azure.ai.textanalytics.linkedentity.url?view=azure-dotnet

N Y Y is correct
First N - there should be additional code shown like "response.doc.primary_language.name". It is not shown (the language detection is available though)
Second Y -
Third Y -
"LinkedEntity contains a link to the well-known recognized entity in text. The link comes from a data source like Wikipedia or Bing. It additionally includes all of the matches of this entity found in the document."

i dont know what image you guys are referring to but this doesnt make sense at all.

NNY - LinkedEntity.Url normally contains a link to wikipedia

No No Yes

While the MS Learn specifically mention Wikipedia, the Class Doc for LinkedEnty refers to Bing. https://learn.microsoft.com/en-us/java/api/com.azure.ai.textanalytics.models.linkedentity?view=azure-java-stable

So Bing answers are correct.
=> NYY

No, the code you provided does not detect the language of documents. It uses the RecognizeLinkedEntities method of the TextAnalyticsClient to identify and link entities in the text to more information on the web.

Yes, the url attribute returned for each linked entity will be a Bing search link. This link provides more information about the entity.

Yes, the matches attribute returned for each linked entity will provide the location in the document where the entity is referenced. It includes the text of the entity, its offset (the number of characters from the beginning of the document to the start of the entity), and its length (the number of characters of the entity).

Final Answer:
N
N
Y

NNY
Tried out and the response does not detect language. it just returns the language code sent in the request and only english and Spanish is supported.
The version 3.1 returns Bing ID however all links returned are only wiki links

The code will detect the language of documents

**No**. The provided code uses the RecognizeLinkedEntities method, which recognizes and classifies named entities in a document and links them to more information on the web.

Yes. The RecognizeLinkedEntities method returns a list of recognized entities linked to more information on the web, typically in the form of a Bing search URL.

Yes. For each linked entity recognized by the RecognizeLinkedEntities method, the matches attribute contains information about occurrences of that entity within the input text.

Incorrect image is displayed for me

N
N
Y
he language that the document is written in. If unspecified, this value will be set to the default language in DefaultLanguage in the request sent to the service. If set to an empty string, the service will apply a model where the language is explicitly set to "None".

Y, N, Y

https://learn.microsoft.com/en-us/dotnet/api/azure.ai.textanalytics.textanalyticsclient.recognizelinkedentities?view=azure-dotnet
Y – language String
The language that the document is written in. If unspecified, this value will be set to the default language in DefaultLanguage in the request sent to the service.
Default language value to use in all client calls. If no value is specified, "en" is set as default.

https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/entities-linking/entities-linking?tabs=HTTP#linkedentity
N – LinkedEntity dataSource string
Data source used to extract entity linking, such as Wiki/Bing etc.

https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/entities-linking/entities-linking?tabs=HTTP#match
Y – Match offset integer
Start position (in Unicode characters) for the entity match text.

Answer is correct.
N - code will not detect language
Y - Data source used to extract entity linking, such as Wiki/Bing etc. URL for the entity's page from the data source.
Y - matches: List of instances this entity appears in the text.
https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/entities-linking/entities-linking?tabs=HTTP

B is correct answer : Mirror Finance to an Azure SQL database.

https://docs.microsoft.com/en-us/azure/search/search-indexer-overview#supported-data-sources

Selected Answer: B

Many similar questions are registered; I'm not sure why only Finance is on-premise.

B is right answer.

A modified version of this was in the exam today.

Agreed is B

appeared in august exam

private endpoint should be used

the answer is B because the traffic should not be routed over internet. "By default, Azure AI Search is configured to allow connections over a public endpoint. Access to a search service through the public endpoint is protected by authentication and authorization protocols, but the endpoint itself is open to the internet at the network layer for data plane requests." https://learn.microsoft.com/en-us/azure/search/service-configure-firewall#when-to-configure-network-access

B seems correct

B is the correct answer.

This answer is no.

B is correct answer : No.

Again NO

Correct Answer is B. No. this scenario routes over public internet, to do this without touching public internet you would use a private endpoint on a vnet then private link to access it.

Vote for 'B'. Private Link is sufficient

SEEMS NO

Yes, is the answer.

Answer is correct.
Instead created private end point and private link.

Use a private endpoint instead of a public endpoint to prevent from routing traffic over the public internet.

appeared in august exam

private endpoint should be used

B is the correct answer.

This answer is no.

Again NO

The answer is B. Even if you configure vnet with a network security group, the app is still routing traffic to the public until it hits the vnet1 where then the network goes secure/private. That still does not address "no public routing of traffic" requirement

SEEMS YES

I think this is the answer based on these two links.
https://docs.microsoft.com/en-us/azure/virtual-network/network-security-groups-overview#network-security-groups

https://docs.microsoft.com/en-us/azure/virtual-network/virtual-networks-overview

Should be ABF (use AKV for customer managed key)

Verified with Azure docs, A,B and F are correct answers.

The provided answers are correct.

"E" should read Query performance will decrease

A, B, F

Coffee gift is ABF.

Based on the information from Microsoft’s documentation12, the implications of enabling server-side encryption and using customer-managed keys (CMK) stored in Azure for Azure Cognitive Search are:

A. The index size will increase. Enabling CMK encryption increases the index size1.
B. Query times will increase. Enabling CMK encryption degrades query performance. You can expect to see an increase of 30-60 percent in query times1.
F. Azure Key Vault is required. CMK encryption depends on Azure Key Vault. You can create your own encryption keys and store them in a key vault, or you can use Azure Key Vault APIs to generate encryption keys1.
So, the correct answers are A, B, and F. Please note that these changes might also lead to additional costs2. It’s always a good idea to review the Azure pricing details before making such changes.

Based on the Azure documentation and best practices for Azure Cognitive Search with server-side encryption using customer-managed keys (CMKs), it appears that the correct answers to the question are B and F. Enabling CMKs is known to potentially increase query times due to the additional encryption/decryption steps required when data is requested (B). Azure Key Vault is required to store and manage these CMKs, making it a necessary component of the configuration (F). There is no indication in the documentation that index size will be affected (A and D), nor that query times will decrease (E), and a self-signed X.509 certificate is not mentioned as a requirement (C).

ABE can't be the right answer. B and E are conflicting

I think this answer highlights why you should come to the comments and do your own research. The answer here says the query time will both increase AND decrease, which doesn't make sense. The answer is ABF, not ABE

The E is wrong! The query time will increase with CMK

Customer-managed keys require another billable service, Azure Key Vault, which can be in a different region, but under the same subscription, as Azure AI Search.

Enabling CMK encryption will increase index size and degrade query performance. Based on observations to date, you can expect to see an increase of 30-60 percent in query times, although actual performance will vary depending on the index definition and types of queries. Because of the negative performance impact, we recommend that you only enable this feature on indexes that really require it.

A,B: "Enabling CMK encryption will increase index size and degrade query performance"
F: "Customer-managed keys require another billable service, Azure Key Vault, which can be in a different region, but under the same subscription, as Azure Cognitive Search."
https://learn.microsoft.com/en-us/azure/search/search-security-overview#data-protection

Enabling CMK encryption will increase index size and degrade query performance. And you need a key vault to store the keys

https://learn.microsoft.com/en-us/azure/search/search-security-overview#data-protection

Index and query will increase because of the encryption.
Because customer wants to manage the keys he will need Azure key vault

ABF seem correct

ABF looks right, B and E are contradicting

The correct answer is D. Use an indexer connected to Azure Cosmos DB that includes language detection and text translation skills. This solution meets the requirement for handling multiple languages and supports content transformation for English, French, and Portuguese by using the text translation skill, which aligns with the technical requirement for multilingual support. Additionally, using Azure Cosmos DB and associated skill sets ensures content security and role-based access control (RBAC), meeting the requirements for content approval and security.

Document extraction skill only works for files. It's less likely to work with CosmoDB.

Is this question checking if I can spot in the table that the wiki content is kept in CosmosDB and not in Blob storage? Is that the skill an AI-engineer should possess? |
Next level is to know that the data in cosmosDB stored for wiki are not documents or images. Just assume based on language skills?

D 100%

This question was in the test I took in August 2024

D is the answer.

It is D and not C. Document extraction skill is not needed because the format of content is text (as stated in the table above)

Is this question still available on May 21, 2024?

must be D

Therefore, A. an indexer for Cosmos DB attached to a skillset that contains the language detection skill and the text translation skill aligns with the technical requirements for content approval, multilingual support, and RBAC1

Must be D

Windows Copilot:
B. An indexer for Azure Blob storage attached to a skillset that contains the language detection skill.

Here’s the explanation:

Azure Blob Storage: You’ll use Azure Blob storage to store your content, including wiki articles and other relevant documents.

Indexer: An indexer in Azure Cognitive Search automatically discovers new content from supported data sources (such as blob storage) and adds it to the search index. In your case, you’ll create an indexer to process the wiki content.

Skillset: A skillset defines a set of cognitive skills (such as language detection, text translation, and more) that can be applied during indexing. Since your knowledgebase needs to support multiple languages (English, French, and Portuguese), you’ll attach a skillset to the indexer.

...

You only need to detect language and translate form wiki

It is confusing because the table mentions wikis are in Text format, but on the other hand, wikis (and blog entries) usually contain HTML.
If they do contain HTML is should be C (extract text from html - dont translate tags)
If they don't it should be D

At the end I choose D because we shouldn't assume stuff, the questions says "Text" not "Text/HTML"

Wikis are already in text format and available in three languages so what do you need the document extraction and translation normalize the text to a single language before indexing for? What you need is to detect the language of the text and translate it to English if you want to normalize the text to a single language before indexing for search. Only in this case it would make sense to use text translation skill.

Language Detection is not needed. The capability is especially useful when you need to provide the language of the text as input to other skills (for example, the Sentiment Analysis skill or Text Split skill). But it is not required for Text Translation

C is correct

Answer is correct. Below link shows the details
https://docs.microsoft.com/en-us/azure/search/knowledge-store-projections-examples

Shinzo ABE!

the skillset definition JSON configuration so tableName, generatedKeyName and source.
ABE is correct on this question in my personal opinion, which belongs to the hand-wringing squadron Turbo Rangers.

For implementing a table projection in a skillset definition for Azure Cognitive Search, the correct properties to specify in the JSON configuration are:

A. tableName: Necessary for defining where the output data should be stored.
B. generatedKeyName: Important for scenarios requiring unique identification of rows, though its necessity can vary.
E. source: Essential for specifying the input data that will be processed and projected into the table.
Therefore, the correct answers are A (tableName), B (generatedKeyName), and E (source), as they play direct roles in the configuration of a table projection within a skillset for Azure Cognitive Search.

correct
https://learn.microsoft.com/en-us/azure/search/knowledge-store-projections-examples#define-a-table-projection

A, B and E are correct answers.

To define a table projection, use the tables array in the projections property. A table projection has three required properties:

tableName: Determines the name of a new table created in Azure Table Storage.

generatedKeyName: Column name for the key that uniquely identifies each row. The value is system-generated. If you omit this property, a column will be created automatically that uses the table name and "key" as the naming convention.

source: A path to a node in an enrichment tree. The node should be a reference to a complex shape that determines which columns are created in the table.

https://docs.microsoft.com/en-us/azure/search/knowledge-store-projections-examples#define-a-table-projection

The answer is wrong
regenerated secondary key
update the app to use the secondary key
regenarated the primary key

The answer is correct. The question specifically mentioned a read-only access to the documents which is achieved only through query keys. And it is a query key that was compromised. Any of the admin keys (primary and secondary) allows read-write operations. Hence, solution is to create a new query key, switch to the new query key and delete the compromised query key.

how can exam topics can be so wrong??

Given that the scenario specifically mentions QUERY key, the answer is correct.

i chose:
regenerated secondary key
update the app to use the secondary key
regenarated the primary key

Since access is stated as read-only the answer is correct

regenerated secondary key
update the app to use the secondary key
regenarated the primary key

The stated answer is correct.
- Query keys do not have the concept of primary and secondary key.
- The concept of primary key and secondary key applies only to admin keys.
- You get one key by default, and then you can create additional keys (up to 50).

It's a little bit confused when I met similar question in the topic 4 question 2. 2 questions have almost similar contents but different answer. Can you guys check it out

1. Add a new query key.
2. Change the app to use the new key.
3. Delete the compromised key.

Suspecting unauthorized access doesn’t give you a reason to interrupt legit users. Also by looking the read only access requirement to the document collections already tells you this is about query keys.

ChatGPT answer : The answer is correct but in a different order.
1. Revoke the compromised query key immediately to prevent further unauthorized access.
2. Create a new query key for the application to use for authorized access.
3. Update the search endpoint and application to use the new query key and ensure that it is properly secured.

The answer is correct. As its about query key and not for admin key.

Answer is correct.
Add a new query key
Change the app to use the new key
Delete the compromised key

1. Regenerate 2nd key
2. Change the app to use the regenerated 2nd key
3. Regenerate the 1st key

Given Answer is Correct:

Add a new query key
Change the app to use the new key
Delete the compromised key

The key here is to not confuse "query Keys" with "admin keys"

https://docs.microsoft.com/en-us/azure/search/search-security-api-keys

The answer is correct. In order to have read only access when you call the search API you need to use query key.

In Azure Blob Storage there's no limit to the size of a single container so that rules out A and B. Also, C is an obvious distractor. So that leaves only D, which makes sense an an answer.

Selected Answer: D

In my opinion, D is correct on this question.

D makes sense. "virtual folders".

FOCUS "virtual folders" word.

Tricky question think of virtual folder AS blob containers and the answer will be obvious

e, option D is the best choice because it leverages the scalability and parallel processing capabilities of Azure Cognitive Search to efficiently index a large volume of documents. By organizing documents into virtual folders and creating an indexer for each folder, you can maximize the throughput of the indexing process. Increasing search units further supports this by allocating more resources to the task, thereby minimizing the time required to make the scanned documents searchable.

"One search unit in your service can run one indexer at any given time. Creating multiple indexers is only useful if they can run in parallel" so A and C are out. B is out as you are not running the indexers in parallel. Besides it's hard to image that with millions of scanned you don't have virtual folders in place to split the data already.

D is correct answer.

Also marked correct on Udemy course practice test.

Was on exam 7 Jun 2022

correct ans

how to do this is defined here:

https://docs.microsoft.com/en-us/azure/search/search-howto-indexing-azure-blob-storage#index-large-datasets
The response is missing the data source creation for each virtual folder or blob container.
D is not correct, but the less wrong of a response…
So I give it a "pass", nowadays it is misleading and not fully correct...

seems to be correct

Create Private Endpoint
Use Azure Roles

1. Private Endpoint
https://learn.microsoft.com/en-us/azure/search/service-create-private-endpoint

2. Key authentication
https://learn.microsoft.com/en-us/azure/search/search-security-api-keys?tabs=portal-use%2Cportal-find%2Cportal-query

API keys control access to the entire service, but do not restrict specific queries. So answer is to use RBAC which is 'Use Azure Roles' for second question

1. Create a private endpoint
2. Use Azure roles

https://learn.microsoft.com/en-us/azure/search/keyless-connections?tabs=csharp%2Cazure-cli

Azure Cognitive Search primarily relies on API keys for authentication and authorization. By generating different API keys, you can control and restrict the access each app has to specific queries. Azure Roles are not used directly for query-level permissions in Azure Cognitive Search.

So, the complete solution would be:

Prevent access to Search1 from the internet: Configure a Private Endpoint for Search1
Limit the access of each app to specific queries: Use key authentication

Create private endpoint
use key auth - basically using query key will restrict the queries to data itself and will not retrieve system level info..

1. Create a private endpoint
2. Use Azure roles

1. Create a private endpoint
2. Use Azure roles

you can configure IP firewall and only allow the apps in. Implementing private endpoint requires the apps to be on the same vnet or a s2s vpn which adds complexity.
https://learn.microsoft.com/en-us/azure/search/service-configure-firewall
https://learn.microsoft.com/en-us/azure/search/search-security-api-keys?tabs=rest-use%2Cportal-find%2Cportal-query

IP Firewall
Use Key authentication.

Create a private endpoint & Use key authentication.

- Private endpoint
- Azure Roles

There is only one Cognitive Search instance; keys will not control access at the correct level of granularity.

Create Private Endpoint
Use Key authentication.


Azure roles, specifically Azure Role-Based Access Control (RBAC), are designed to manage who has access to Azure resources and what they can do with those resources. While RBAC is effective for controlling access at the resource level (e.g., who can manage the search service, indexes, or data sources), it does not provide the granularity needed to limit access to specific queries within Azure Cognitive Search.

Final Answer:
1. Create a private endpoint
2. Use Azure roles

In this article, learn how to secure an Azure AI Search service so that it can't be accessed over a public internet connection:

https://learn.microsoft.com/en-us/azure/search/service-create-private-endpoint

prevent Azure services to be accessed from internet? ==>create private endpoint of that service
limit the service to a specific queries?-->using Azure roles

Should be
1. Create Private Endpoint
2. Use Azure Roles

for me is B, because add partition is the correct solution

The answer is A. According Microsoft materials:
A simple fix to most throttling issues is to throw more resources at the search service (typically replicas for query-based throttling, or partitions for indexing-based throttling).

I think the answer should be B.
Aditional replicas can improve query latency. However, you can allocate up to 12 replicas in a Standard search service and 3 replicas in a Basic search service. As there is no information about the actual number of configured replicas, I think upgrading the princing tier is the only sure solution because it comes with more cpu and memory for each replica.

"add replicas" is correct.

Some question sets might have more than one correct solution,so A.

The correct answer is B. No.

Here’s why:

Adding replicas does not directly address the issue of query throttling in Azure Cognitive Search.
Query throttling occurs when the service limits the number of requests it can handle within a given time frame. It is typically related to the service’s capacity, rate limits, and resource availability.
To reduce the likelihood of query throttling, consider other factors such as:
Service capacity: Ensure that your service tier (including replicas and partitions) can handle the query volume.
Optimize query design: Efficiently structure your queries to minimize resource usage.
Index composition: Smaller indexes tend to perform better, but also consider schema complexity and field attribution.
Therefore, adding replicas alone does not meet the goal of reducing query throttling. You need to address other aspects of your search service to achieve better performance and avoid throttling

A is right.

According to Microsoft's training material:

**Use the best service tier for your search needs**
You've seen that you can scale out service tiers by adding more partitions. You can scale out with replicas if you need to scale because of an increase in load. You can also scale up your search service by using a higher tier.

add replicas can help read performance;
write performance needs adding partittions

The answer is correct.

Quote "In Cognitive Search, replicas are copies of your index." at https://learn.microsoft.com/en-us/azure/search/search-reliability

From : https://learn.microsoft.com/en-us/azure/search/search-performance-tips#index-size-and-schema

However, if the index is right-sized, the only other calibration you can make is to increase capacity: either by adding replicas or upgrading the service tier

So adding replicas and upgrading the sku is two valid answers for this question

Answer is correct

Shouldn't it be Cognitive Search with a higher tier?

I wrote this exam on 25 Dec 2024
No simulation
and this was one for the question I got in Case study.

Score 805 (all I did was study this and revise from number 1 to 333 for 6hours) memorize the case study question and answers too.

In all Pray for God's direction.
Jesus Christ is Lord and our guide

This question was in the test I took in August 2024

A is right answer.
From Takedajuku perspective, if you study for 4 days and spend 2 days reviewing, you will have a better chance of passing the exam.

Is this question still available on May 21, 2024?

A is correct

Quote "In Azure Cognitive Search, a synonym map contains a list of rules for expanding or rewriting a search query to equivalent terms." at https://learn.microsoft.com/en-us/rest/api/searchservice/create-synonym-map .

A is correct answer.

Synonym maps are a global resource that associate equivalent terms, expanding the scope of a query without the user having to actually provide the term.
You might create multiple synonym maps for different languages, such as English and French versions, or lexicons if your content includes technical or obscure terminology.
https://docs.microsoft.com/en-us/azure/search/search-synonyms