AI-103 Information Extraction — Timed 20 Questions

Question 1

You are troubleshooting an Azure AI Document Intelligence custom extraction model. When you test a new document, certain currency values are being extracted with the currency symbol missing. Where should you look in the JSON response structure to find the normalized value?

20 minute time limit — choose an answer to begin.

Accepted Answer

In the documentFields array under the valueCurrency object property.. In Document Intelligence's analyze response, extracted fields are organized under the documents array, and each field has a typed value object. For a currency field, the normalized value (amount plus currency code) is exposed under valueCurrency, which includes amount and currencyCode properties. The raw text with the symbol may appear in the field's content, but the normalized representation lives in valueCurrency.

Answer

In the styles collection under the isHandwritten property.

Answer

In the paragraphs collection under the boundingRegions array.

Answer

In the analyzeResults object under the content property.

Question 2

You are designing a high-throughput document processing pipeline using Azure AI Document Intelligence. To optimize latency and cost, you want to extract only the text and layout structures (paragraphs, lines, words, and tables) from a multi-page PDF without performing key-value extraction or using custom models. Which model ID should you specify in your API request?

Accepted Answer

prebuilt-layout. prebuilt-layout extracts text plus layout structures — paragraphs, lines, words, selection marks, and tables — without key-value pairs or custom training. It is the correct model when you need structural document understanding at scale. prebuilt-read returns only raw text, and prebuilt-document adds key-value extraction, which the scenario explicitly excludes.

Answer

prebuilt-invoice

Answer

prebuilt-read

Answer

prebuilt-document

Question 3

You are configuring a custom extraction model in Azure AI Document Intelligence and want to compose multiple specialized models into a single endpoint that automatically classifies and routes documents. Which REST API endpoint should you use to create this composed model?

Accepted Answer

POST /documentintelligence/documentModels:compose?api-version=2024-02-29-preview. The compose operation is exposed as POST /documentintelligence/documentModels:compose?api-version=2024-02-29-preview, which takes a list of existing custom model IDs and a new modelId, and returns a composed model that can classify and route documents to the appropriate submodel. This is the documented endpoint for building composed models.

Answer

PUT /documentintelligence/documentModels/{modelId}?api-version=2024-02-29-preview

Answer

POST /documentintelligence/documentModels:train?api-version=2024-02-29-preview

Answer

POST /documentintelligence/models:classify?api-version=2024-02-29-preview

Question 4

You are building an automated invoice processing pipeline using Azure AI Document Intelligence. Some incoming invoices contain handwritten vendor signatures and handwritten approval dates. Which feature must you ensure is supported and enabled when configuring your analysis request?

Accepted Answer

Rely on the built-in OCR capability which detects both printed and handwritten text across supported models.. Azure AI Document Intelligence's built-in OCR (Read model) natively detects both printed and handwritten text across its supported prebuilt and custom models, so no special parameter or separate service is required. Handwriting recognition is a core capability of the underlying Read/OCR engine, meaning the invoice model can extract handwritten signatures and approval dates without additional configuration beyond ensuring the model supports it. This makes option A the correct answer because it reflects the actual architecture of the service.

Answer

Enable the 'handwritten=true' query parameter in the API call header.

Answer

Train a custom neural model exclusively with handwritten signature samples.

Answer

Deploy a separate Azure AI Vision Handwritten Text container and pipe its output into the Document Intelligence request.

Question 5

Which TWO prebuilt models are officially available in Azure AI Document Intelligence for extracting specialized domain data? (Choose TWO)

Accepted Answer

prebuilt-tax.us.w2. Option C, prebuilt-tax.us.w2, is correct because Azure AI Document Intelligence ships a dedicated US W-2 tax form model that extracts fields such as wages, tips, and federal income tax withheld from IRS W-2 documents. Option E, prebuilt-invoice, is correct because the prebuilt invoice model is an officially supported specialized-domain model that extracts key invoice fields like vendor name, invoice date, due date, and line items. The other options are not official prebuilt model IDs: prebuilt-socialSecurityCard, prebuilt-architecturalDrawing, and prebuilt-medicalChart do not exist in the Document Intelligence prebuilt model catalog, so they cannot be selected.

Answer

prebuilt-socialSecurityCard

Answer

prebuilt-architecturalDrawing

Answer

prebuilt-medicalChart

Question 6

You are configuring an Azure AI Document Intelligence custom neural model training run. Which TWO prerequisites must be met regarding your training dataset? (Choose TWO)

Accepted Answer

Training documents must be stored in an Azure Blob Storage container with a SAS URI or managed identity access.. Option A is correct because Azure AI Document Intelligence custom neural model training requires the training documents to reside in an Azure Blob Storage container, and the service accesses them either through a SAS URI (shared access signature) or via managed identity authentication. Option B is correct because the documented minimum for custom neural model training is at least five training documents per field or form type, ensuring the model has enough labeled examples to learn the field's layout and semantics. Option C is incorrect because Document Intelligence accepts PDF, TIFF, JPG, PNG, BMP, and other supported formats; .docx conversion is not required. Option D is incorrect because labels can be generated through Document Intelligence Studio's labeling interface, and the labels.json (or .ocr.json/.labels.json) files are produced automatically rather than requiring a hand-written Python script. Option E is incorrect because there is no requirement that all training documents share identical pixel dimensions or DPI settings; the service normalizes and processes varied inputs.

Answer

All training documents must be converted to Microsoft Word (.docx) format prior to upload.

Answer

You must manually write a Python script to generate the labels.json file instead of using Document Intelligence Studio.

Answer

Training documents must all have identical pixel dimensions and DPI settings.

Question 7

Which THREE roles or permissions are typically required to successfully create, train, and manage custom models in Azure AI Document Intelligence Studio? (Choose THREE)

Accepted Answer

Cognitive Services User to invoke model prediction APIs. Option B (Cognitive Services User) is correct because invoking the trained custom model's prediction/analysis APIs requires the Cognitive Services User role, which grants the data-plane permission to call the Document Intelligence endpoint. Option C (Storage Blob Data Contributor) is correct because Document Intelligence Studio reads the labeled training dataset from Azure Blob Storage, and the signed-in user needs data-plane access to that container to build and train the model. Option E (Cognitive Services Contributor) is correct because creating, training, and managing custom models requires management-plane control over the Document Intelligence resource, which the Cognitive Services Contributor role provides. Option A (Global Administrator) is not required, since tenant-wide admin rights are unnecessary and violate least privilege for this task. Option D (Network Contributor) is not required unless private networking is being configured, which is not part of the standard create/train/manage workflow.

Answer

Global Administrator on the Microsoft Entra ID tenant

Answer

Network Contributor on the subscription virtual network

Question 8

Your company processes complex purchase orders that vary significantly in layout across different vendors. You decide to train a Custom Neural model using Azure AI Document Intelligence Studio. What is the minimum number of documents required to train a Custom Neural model?

Accepted Answer

5 documents. Azure AI Document Intelligence Custom Neural models require a minimum of 5 labeled documents to begin training. This low threshold exists because neural models leverage pretrained language understanding and can generalize from a small labeled sample, unlike template-based custom models that need more examples to lock onto fixed layouts. The 5-document minimum applies per document type being trained.

Answer

1 document

Answer

50 documents

Answer

100 documents

Question 9

You are training a Custom Template model using Azure AI Document Intelligence. Your training dataset contains documents with multiple pages where the tables span across page breaks. How should you structure your training labels to handle multi-page tables correctly?

Accepted Answer

Label the table rows across pages using the same table index and consistent headers so the model recognizes the continuation.. Azure AI Document Intelligence Custom Template models support multi-page tables by allowing you to label rows across page breaks using the same table index and consistent headers. This signals to the model that the table continues on subsequent pages, enabling it to correctly extract the entire table as a single entity. The labeling tool (e.g., Document Intelligence Studio) automatically handles page breaks when you assign the same table index to rows on different pages, provided the headers match. This approach ensures the trained model can recognize and extract multi-page tables without post-processing.

Answer

Create a separate table label for each page fragment and concatenate them via post-processing code.

Answer

Exclude multi-page tables from the training set and rely solely on the prebuilt layout model for tables.

Answer

Convert all multi-page documents into single extremely long virtual pages before labeling.

Question 10

When composing multiple individual Custom Neural models into a single composed model in Azure AI Document Intelligence, which THREE behaviors or characteristics apply? Each correct selection represents a valid aspect.

Accepted Answer

All sub-models included in a composed model must share the same major API version.. Option B is correct because all sub-models composed together must be built with the same major API version (for example, all 2023-07-31 or all 2024-02-29-preview); mixing major versions is not supported. Option D is correct because a composed model performs document-type classification first, then routes each incoming document to the matching custom sub-model, so callers do not have to select the model themselves. Option E is correct because the composed model is addressed by one model ID, so client code calls a single AnalyzeDocument endpoint/model rather than branching per document type. Option A is wrong because all sub-models and the composed model live in the same Azure AI Document Intelligence resource and use that resource's single endpoint. Option C is wrong because there are practical limits on the number of sub-models per composed model and composition adds classification latency, so it is not unlimited or performance-neutral.

Answer

Composed models require a separate Azure resource endpoint for each individual sub-model.

Answer

Composed models support an unlimited number of sub-models without any performance impact.

Question 11

Which TWO of the following are prerequisites for training a custom extraction model?

Accepted Answer

A set of labeled data in JSON format. Option A is correct because training a custom extraction model (such as Azure AI Document Intelligence's custom extraction model) requires a labeled dataset in JSON format that specifies the fields to extract and their values, typically produced via the Document Intelligence Studio labeling tool. Option E is correct because the training documents themselves must be stored in an Azure Blob Storage container, which the service accesses as the source of the training data. Option B is not required, since Azure SQL Database is not used to supply training data for custom extraction models. Option C is not required, as a pre-existing template model is not a prerequisite for training a custom extraction model. Option D is not required, because labeling is performed through the cloud-based Document Intelligence Studio rather than a pre-installed local labeling agent.

Answer

An Azure SQL database

Answer

A pre-existing template model

Answer

A pre-installed local labeling agent

Question 12

Which THREE file types are supported for direct processing by the Document Intelligence API?

Accepted Answer

TIFF. The Document Intelligence API (formerly Form Recognizer) accepts TIFF (B), JPEG (C), and PDF (D) as supported input formats for direct processing, since these are the image and PDF types the service's OCR and layout models can ingest natively. TIFF and JPEG are supported because the service performs optical character recognition on raster images, and PDF is supported because it can process both text-based and scanned PDF documents directly. DOCX (A) and PPTX (E) are not directly supported input formats; Office files must first be converted to PDF or an image format before submission to the API.

Answer

DOCX

Answer

PPTX

Question 13

Which THREE items are included in the JSON response from the Document Intelligence analyze operation?

Accepted Answer

PageResults (structure). The Document Intelligence (formerly Form Recognizer) analyze operation returns a JSON payload whose top-level sections include ReadResults (B), which holds the OCR output — lines, words, and their bounding boxes per page; PageResults (A), which contains per-page structural information such as page dimensions, angle, and the selection marks (checkboxes) detected; and DocumentResults (E), which contains the extracted key-value pairs and field confidence scores for each document when a model with field extraction (e.g., prebuilt invoice/receipt or custom model) is used. These three sections are the standard components of the analyze response, so A, B, and E are correct. UserPermissions (C) is not part of the analyze response — access control is handled via Azure RBAC/keys, not returned in the result payload. TrainingLogs (D) are not included in the analyze response either; training diagnostics belong to model-training operations (e.g., the build/custom model APIs), not to the analyze result.

Answer

UserPermissions

Answer

TrainingLogs

Question 14

You are using Azure AI Document Intelligence to extract data from a set of purchase orders. The documents have a consistent layout, but the training dataset contains only five labeled samples. You need to train a custom model that can extract fields such as order number, vendor name, and total amount. Which model type should you choose to achieve the best accuracy with minimal samples?

Accepted Answer

Custom template model. Custom template models are optimized for documents with consistent layouts and can be trained with as few as five labeled samples. They use a template-based approach that matches field positions and labels, providing high accuracy for structured documents like purchase orders with a fixed layout. This makes them the ideal choice when training data is limited and layout variability is low.

Answer

Prebuilt invoice model

Answer

General document model with custom field extraction

Answer

Custom neural model

Question 15

You are building a .NET console application that uses the Azure AI Document Intelligence SDK to analyze receipts. The application must submit a local PDF file for analysis using the prebuilt receipt model and then retrieve the extracted fields. You need to write the minimal code that initiates the analysis and returns the result. Which SDK method should you call on the DocumentAnalysisClient to start the analysis and obtain an operation handle?

Accepted Answer

StartAnalyzeDocumentAsync. The current Azure AI Document Intelligence .NET SDK exposes DocumentAnalysisClient with StartAnalyzeDocumentAsync for stream-based input and StartAnalyzeDocumentFromUriAsync for URL input. Because the scenario submits a local PDF, the stream-based start method is correct. It returns an operation that must be polled to completion before reading the AnalyzeResult.

Answer

BeginAnalyzeDocumentAsync

Answer

StartAnalyzeDocumentFromUriAsync

Answer

AnalyzeDocumentAsync

Question 16

You are developing an application that uses Azure AI Document Intelligence to extract data from receipts. You need to ensure that the extracted data includes the merchant name, transaction date, and total amount. You also want to handle documents where the receipt is rotated or skewed. Which TWO actions should you perform? (Choose two.)

Accepted Answer

Ensure the input documents are in a supported format such as JPEG or PNG.. The prebuilt receipt model is designed to extract merchant name, transaction date, and total amount from receipts with high accuracy. Additionally, ensuring input documents are in supported formats like JPEG or PNG allows the service to perform automatic rotation and skew correction during OCR. Together, these actions meet the requirements for field extraction and handling of rotated or skewed receipts.

Answer

Set the 'pages' parameter to include all pages in the receipt.

Answer

Enable the 'ocr.rotationCorrection' option in the analyze request.

Answer

Use the general document model and then parse the fields manually.

Question 17

You are using Azure AI Document Intelligence to extract data from a batch of invoices. You need to retrieve the extracted fields such as invoice ID, customer name, and total amount. Which API endpoint should you call to get the analysis results after submitting the document?

Accepted Answer

GET /documentModels/{modelId}/analyzeResults/{resultId}. After submitting a document to the analyze endpoint, the service returns a result ID. To retrieve the extracted fields, you must call the GET /documentModels/{modelId}/analyzeResults/{resultId} endpoint. This returns the analysis results, including invoice ID, customer name, total amount, and confidence scores. It is the correct way to poll for and obtain the results in the asynchronous API workflow.

Answer

GET /documentModels/{modelId}/analyze

Answer

POST /documentModels/{modelId}/analyzeResults

Answer

POST /documentModels/{modelId}/analyze

Question 18

You are developing a custom extraction model using Azure AI Document Intelligence Studio. Your training dataset consists of multi-page invoices with varying structures, and some fields appear only on the final page. How should you structure your tagging configuration?

Accepted Answer

Tag the fields on whatever page they appear across the sample documents in Document Intelligence Studio.. In Document Intelligence Studio, you tag fields on the pages where they actually appear across your sample documents — the labeling tool is page-aware and stores the page number alongside each field's bounding region. This lets the trained model learn that, for example, 'Total' appears on page 1 in some invoices and page 3 in others. You do not need to split documents or build orchestration models just because fields appear on different pages.

Answer

Create a separate custom model for the final page and chain them using an orchestration model.

Answer

Split every multi-page invoice into individual single-page documents before uploading.

Answer

Define all late-appearing fields as optional metadata parameters in the training configuration JSON file.

Question 19

You are configuring an Azure AI Document Intelligence orchestration model to automatically classify and extract data from incoming customer correspondence, which includes invoices, work orders, and purchase orders. One of the component models is failing to classify documents correctly. How should you resolve this issue?

Accepted Answer

Retrain the failing component custom model with more distinct sample documents that represent the specific document class.. The most effective way to resolve misclassification by a component model in an orchestration model is to retrain that custom model with more distinct sample documents that clearly represent the specific document class. This improves the model's ability to differentiate between classes.

Answer

Modify the orchestration model routing threshold in the Azure portal settings.

Answer

Switch the orchestration model from a template-based classifier to a neural classifier using the REST API.

Answer

Update the API version to bypass classification and force direct extraction via the general document model.

Question 20

Your team needs to extract data from a large volume of standardized tax forms using Azure AI Document Intelligence. The forms have a fixed layout where fields are always in the exact same location. Which model type provides the fastest training and highest accuracy for this specific use case?

Accepted Answer

Custom template model. A custom template model is designed for documents with a consistent, fixed layout, and it trains quickly because it relies on layout/template matching rather than deep semantic learning. For standardized tax forms where fields are always in the same location, the template model achieves high accuracy with minimal training data and fast training time. This is the classic use case Microsoft documents for the template model type.

Answer

Prebuilt receipt model

Answer

Custom neural model

Answer

General document model