AI-103 · Practice set 6 of 13
Foundry Tools or a Model?: 10 practice questions
10 free AI-103 practice questions on Foundry Tools or a Model?, with an explanation for every answer. Untimed. The full mock exam and the timed version are in the app.
-
Question 1 of 10
An app must transcribe recorded support calls and translate the speech at the same time, without training any model. Which Foundry Tool fits?
- ASpeech
- BContent Safety
- CAzure Vision
- DCustom Vision
Show the answer
Match the input type to the tool first. Audio goes to Speech, which covers speech to text, text to speech, translation, and speaker recognition. Out of the box, you send an audio file and get a transcription and a translation back, with no customization.
Next → 1 / 10 -
Question 2 of 10
A team building a new FAQ bot proposes QnA Maker because a sister team still runs it. What's the right call?
- APick another tool, since it's being retired
- BUse it, because existing apps prove it's supported
- CUse it on the F0 free tier to avoid any risk
- DUse it in an on-premises container for new apps
Show the answer
A tool that still runs for someone else isn't a safe pick for a new design. QnA Maker is on the retirement list: still available for existing applications, but the guidance is not to use it for new AI applications. The same applies to LUIS, Personalizer, and the rest of that list.
Next → 2 / 10 -
Question 3 of 10
A team processes standard invoices and receipts that follow common templates. Which option matches that scenario?
- AThe CU prebuilt-contract analyzer
- BA DI container in an air-gapped site
- CA model you prompt and chain on your own
- DA Document Intelligence prebuilt model
Show the answer
Standard structured forms such as invoices, receipts, IDs, tax, or mortgage documents match a Document Intelligence prebuilt model. It's built for structured documents where consistency, low latency, and proven accuracy come first.
Next → 3 / 10 -
Question 4 of 10
A team only needs the text and page layout of scanned pages, at the lowest cost. Which Content Understanding option fits?
- AA DI custom model with 5 labeled samples
- BThe CU prebuilt-contract analyzer
- CA CU custom analyzer with a knowledge source
- DCU prebuilt-read or prebuilt-layout
Show the answer
When the job is plain OCR or layout, don't pay for extraction you won't use. CU prebuilt-read or prebuilt-layout gives lower cost and richer layout extraction than training or customizing anything.
Next → 4 / 10 -
Question 5 of 10
Contracts state a start date and a duration but never the end date. Which option is recommended to infer that date?
- AA Content Understanding analyzer
- BA Document Intelligence prebuilt model
- CLayout with CU prebuilt-layout
- DOCR with CU prebuilt-read
Show the answer
The end date isn't written on the page, so the task needs inference. Content Understanding can infer a field from other values, such as an end date from a start date and duration. Plain text or layout extraction only returns content already present.
Next → 5 / 10 -
Keep the ones you got wrong
In the app, every question you miss comes back exactly when you’re about to forget it.
-
Question 6 of 10
Compared with Content Understanding, what does a custom solution built directly on a Foundry model lack until you implement it?
- AControl over model selection
- BIntegration with your own AI stack
- CConfidence scores and grounding
- DOwnership of the prompts it uses
Show the answer
The managed tools return confidence and grounding with each result. A custom build has none by default, so you either accept or review everything, or you build your own confidence model. The other options are the reasons people build their own.
Next → 6 / 10 -
Question 7 of 10
Highly structured claim forms need custom fields, and the team can label examples. Which tool fits, and how much labeled data does it need to start?
- AA DI custom model, with as few as 5 labeled samples
- BA CU custom analyzer, with at least 50 labeled samples
- CThe CU prebuilt-contract analyzer, with no labels
- DThe CU prebuilt-layout analyzer, with no samples at all
Show the answer
Labels plus highly structured documents point to a Document Intelligence custom model, whose neural training starts with as few as 5 labeled samples. A CU custom analyzer is the choice when you have no labels and describe fields in plain language instead.
Next → 7 / 10 -
Question 8 of 10
A hospital must extract fields from intake forms at a site with no internet connection, and the data can't leave the building. The forms have a few known variants. Which option fits?
- AA CU custom analyzer, zero-shot
- BDocument Intelligence containers
- CA DI prebuilt model via the cloud API
- DA CU prebuilt analyzer for variants
Show the answer
The deployment constraint outranks the document type. For on-premises or air-gapped deployment, Document Intelligence containers are the only option today. A CU analyzer would otherwise suit known variants well, but it doesn't meet the disconnected requirement.
Next → 8 / 10 -
Question 9 of 10
Invoices arrive from hundreds of vendors with inconsistent layouts. The team has no labeled data, needs fields no prebuilt covers, and wants accuracy to improve over time. Which approach fits?
- AA CU custom analyzer, zero-shot, adding labels later
- BA model with prompts you chain, plus your own confidence model
- CA DI custom model with five samples of each vendor's layout
- DThe CU prebuilt-contract analyzer, applied with no setup
Show the answer
Three constraints decide it: no labels, high template variation, and custom fields. A CU custom analyzer starts zero-shot from plain-language field descriptions, generalizes across variation, and improves when you add labeled samples to its knowledge source.
Next → 9 / 10 -
Question 10 of 10
A team must use its own fine-grained prompts and plug extraction into a proprietary AI platform that Content Understanding can't accommodate. It accepts building its own confidence scoring. Which approach fits?
- ACU prebuilt-layout plus rules
- BA DI custom model with labels
- CBuild directly on Foundry models
- DA CU custom analyzer in zero-shot mode
Show the answer
This is the exception case. Complete control over model selection, fine-grained prompt ownership, or a proprietary AI infrastructure that Content Understanding can't accommodate make building directly on Foundry models appropriate, at the price of handling OCR, scaling, and confidence yourself.
Next → 10 / 10 -
You’ve finished this set
That’s 10 questions on Foundry Tools or a Model?. In the app the ones you miss come back exactly when you’re about to forget them.
The whole course, on your phone
Lessons you can read, audio you can listen to on the way to work, and practice that remembers what you got wrong.
“This app is the ‘One Lake’ for preparing the certification exams. Really worth it.”