UiPath Specialized AI Professional (UiSAIv1) — Free Practice Questions
10 free sample questions from a bank of 121, with the correct answers and explanations. No signup required — start practising right now.
1When is it recommended to use Main-ActionCenter in the context of the Document Understanding Process?
When implementing an attended process.
When testing locally or implementing an attended process.
When testing locally.
When testing locally or implementing an unattended process.
Answer: D
The short version
D — Local testing plus unattended Action Center validation is the intended split. Main-ActionCenter supports local debugging and unattended human validation through Orchestrator queues and Action Center, while attended flows use local Present activities instead.
Key concepts in this question
Main-ActionCenter template: the Document Understanding process variant wired for Orchestrator queues plus Create and Wait for validation actions
Attended versus unattended validation: attended uses Present Validation Station on the user machine, unattended suspends and resumes through Action Center
Local testing: Studio debugging of the same unattended template before publishing to Orchestrator
Document Understanding Process: the standard framework with digitize, classify, extract, validate and export stages
Why D is correct
D is correct because UiPath designates Main-ActionCenter for two situations, testing locally and running unattended with Action Center. In local runs it lets developers exercise digitization through export end to end, and in production unattended runs it creates Document Validation actions, suspends the workflow, then resumes after a human acts in Action Center. That dual purpose matches template guidance and separates it from the attended Main path.
Why the others are wrong
A. Attended processes should use the attended Main flow with Present Validation Station, not the suspend and resume Action Center flow
B. Attended use is wrong for Main-ActionCenter, since attended validation stays local and does not need Orchestrator actions or persistence
C. Testing locally alone is incomplete, because it omits the primary production purpose, which is unattended validation through Action Center
UiSAIv1 exam tip
Remember Action in the name means Action Center, which means unattended plus local test, while Present in the name means attended on the local machine.
2In UiPath Communications Mining, when creating a model that is intended to enable automation, what is the minimum F1 score recommended for entities?
60%
70%
80%
90% and above
Answer: D
The short version
D — Community answer is correct. Automation bound Communications Mining models need Excellent rating, which maps to 90 and above, so entity F1 must reach that bar before production use.
Key concepts in this question
Model rating bands: Communications Mining rates models Poor, Average, Good from 70 up, and Excellent from 90 up based on holistic factors.
Entity F1 role: F1 blends precision and recall for extraction fields, showing whether extracted values are reliable enough to drive automation.
Automation risk: Low scoring entities cause wrong downstream actions, so UiPath sets a higher bar for straight through processing than for analytics.
Live data testing: Even with Excellent rating, UiPath recommends testing on live data before deploying automation to production.
Why D is correct
UiPath docs state a score of 70 or above is classed as Good whilst 90 is required for Excellent, and explicitly recommend an Excellent rating when creating a model intended to enable automations. Entities drive the automated values, so the minimum recommended F1 for automation use is therefore 90 percent and above. Option D is the only choice that satisfies the Excellent threshold, while lower scores remain suitable only for insight or semi automated review scenarios.
Why the others are wrong
A. 60 percent is below Good and indicates weak extraction quality that cannot safely drive automation.
B. 70 percent reaches Good but docs still require Excellent for automation, so it is insufficient for unattended use.
C. 80 percent remains in Good band and falls short of the 90 Excellent bar required for automation enablement.
UiSAIv1 exam tip
Link automation to Excellent to 90, Good to 70 is not enough, recall Excellent equals 90 plus for Communications Mining automation questions.
3What type of error does the red exclamation mark in Validation Station or Action Center throw?
DocumentRejectedByUserException.
RejectedException.
DocumentException.
DocumentRejectedByReviewerException.
Answer: A
The short version
A — Community answer is correct. The red exclamation path for a reviewer rejected document raises DocumentRejectedByUserException, which the DU framework catches as a business rule style exception.
Key concepts in this question
Validation Station reject flow: Present Validation Station and Action Center let a reviewer confirm fields or report the document as an exception with a reason.
DU exception taxonomy: UiPath IntelligentOCR Exceptions include BusinessRuleException and DocumentRejectedByUserException for human rejected docs.
Red exclamation meaning: The indicator signals a validation failure or explicit rejection that surfaces as a catchable exception for SetTransactionStatus handling.
Action Center parity: Attended Validation Station and Document Validation actions share the same validation and report exception behaviour.
Why A is correct
DocumentRejectedByUserException was introduced for user rejected docs during validation. Release notes and DU code check TypeOf exception Is DocumentRejectedByUserException alongside BusinessRuleException in SetTransactionStatus. The red exclamation path therefore throws A, which logic treats like a business exception and routes to rejected handling.
Why the others are wrong
B. RejectedException is not the documented DU validation exception name and does not appear in the IntelligentOCR exceptions namespace.
C. DocumentException is too generic and is not the specific type raised for explicit user rejection in Validation Station.
D. DocumentRejectedByReviewerException is not a real UiPath exception name, it confuses reviewer wording with the actual User suffix used by UiPath.
UiSAIv1 exam tip
Remember User not Reviewer for rejection, link red exclamation in Validation Station to DocumentRejectedByUserException and BusinessRuleException handling in the DU template.
4How is the Thread Duration calculated in UiPath Communications Mining's thread analytics?
It is calculated by taking the difference between the timestamps of the first message and the last message sent by the same participant within a thread.
It is calculated by counting the total number of messages in a thread and multiplying it by a fixed amount of time.
It is calculated by taking the time between the first and the last message that was sent in the thread.
It is calculated by taking the difference between the timestamps of the first message and the second message sent in the thread.
Answer: C
The short version
C — Thread Duration is first message to last message in the thread. Communications Mining measures the elapsed wall-clock span across the whole thread, not per participant or per message count, giving a direct handle-time style duration for reports.
Key concepts in this question
Thread analytics: Communications Mining reports that aggregate messages grouped into one conversation thread
Thread Duration: the elapsed time metric shown in thread reports for service time analysis
First to last timestamp: simple max timestamp minus min timestamp within the thread boundary
Using reports docs: the official thread reporting page that defines duration as first to last message time
Why C is correct
C is correct because the documented definition takes the time between the first and the last message sent in the thread, regardless of sender. That spans customer plus agent turns and idle gaps, so it reflects total thread lifetime. It does not filter to one participant and does not multiply message counts by a constant, which would hide real waiting time and distort throughput analysis in Trends and Threads views.
Why the others are wrong
A. Same-participant first to last is wrong, because duration must cover all participants to show full resolution span
B. Count times fixed time is wrong, because it invents a duration instead of measuring actual timestamps and ignores gaps
D. First to second message is wrong, because it captures only initial response lag, not the complete thread lifetime
UiSAIv1 exam tip
For duration questions, think lifetime equals last minus first across everyone, while response lag questions narrow to specific hops.
5What is the relationship between AI Center and UiPath Document Understanding?
AI Center is the infrastructure on top of which UiPath Document Understanding digitization runs.
AI Center is the infrastructure on top of which UiPath Document Understanding machine learning models run.
Document Understanding is the infrastructure on which AI Center machine learning models run.
Document Understanding is the infrastructure on which AI Center digitization runs.
Answer: B
The short version
B — AI Center hosts the Document Understanding ML models. Document Understanding provides the digitize, classify and extract framework, while AI Center supplies ML packages, pipelines, datasets and deployed ML Skills that serve predictions to robots.
Key concepts in this question
AI Center: UiPath ML lifecycle hub for datasets, pipelines, ML packages and ML Skills deployment
Document Understanding: the document pipeline that calls ML models for classification and extraction
ML Skill: a deployed model endpoint consumed by the Machine Learning Extractor and Classifier activities
Digitization versus inference: OCR digitization prepares text, while ML models perform understanding on top
Why B is correct
B is correct because the official relation page states AI Center is the infrastructure on which Document Understanding machine learning models run. Teams label documents in Document Manager, export to AI Center datasets, run training or evaluation pipelines on DocumentUnderstanding ML packages, then deploy an ML Skill referenced by Studio activities. Digitization itself runs in the workflow, but model training, versioning and serving live in AI Center.
Why the others are wrong
A. Digitization running on AI Center is wrong, because digitization is OCR plus DOM creation in the workflow, not an AI Center hosted model
C. Reversed dependency is wrong, because AI Center does not run on Document Understanding, it is the broader ML platform underneath
D. Reversed digitization claim is wrong for the same reason, swapping host and consumer and misplacing OCR work
UiSAIv1 exam tip
Link AI Center to models and ML Skills, and Document Understanding to the consuming pipeline that calls those skills.
6Which validation checks are performed for ML packages uploaded with the Enable Training option inactive?
Existence of a non-empty root folder, requirements.txt file, and main.py file in the root folder which implements a class Main.
The class is further validated to implement an __init__ and a predict function.
Existence of a requirements.txt file, and main.py file which implements a class Main.
The class is further validated to implement an __init__ and a predict function.
Existence of a non-empty root folder, main.py file in the root folder which implements a class Main.
The class is further validated to implement an __init__ and a predict function.
Existence of a non-empty root folder, requirements.txt file, and train.py file in the root folder which implements a class Main.
The class is further validated to implement an __init__ function.
Answer: A
The short version
A — Inference-only packages need root folder, requirements and Main with init plus predict. With Enable Training inactive the package only serves predictions, so validation checks the serving contract, not training entry points or extra scripts.
Key concepts in this question
Enable Training inactive: upload mode for inference-only ML packages that will not run training pipelines
requirements.txt: dependency manifest needed to build the serving container environment
main.py with class Main: the required serving contract exposing model load and inference methods
Init plus predict: constructor loads artifacts, predict takes input and returns extraction or classification output
Why A is correct
A is correct because the managing ML packages docs require a non-empty root folder plus requirements.txt plus main.py in root implementing class Main with __init__ and predict. __init__ loads the model, predict serves it, and requirements ensures reproducible install. Training hooks like train are not required when training is disabled, and the checks intentionally omit train.py to keep inference packages lightweight yet deployable as ML Skills.
Why the others are wrong
B. Omitting non-empty root folder is wrong, because validation explicitly requires a valid packaged root, not loose files
C. Omitting requirements.txt is wrong, because dependencies must be declared for the serving container to install correctly
D. Requiring train.py with only init is wrong, because that describes a training-capable package, not an inference-only upload
UiSAIv1 exam tip
No training means no train function, just init plus predict with requirements in a non-empty root.
7What does the feature in the red highlighted box indicate?
A label annotation.
A label's confidence level.
A label prediction.
A label suggestion.
Answer: D
The short version
D — The highlight shows a label suggestion awaiting confirmation. In Data Manager suggestions appear as model-proposed boxes distinct from human-applied annotations, letting labelers accept or correct them to speed up training data creation.
Key concepts in this question
Data Manager labeling: the web app where users annotate documents to build Document Understanding datasets
Label suggestion: a model-generated candidate box and field assignment shown for review
Label annotation: a human-confirmed box that becomes ground truth after acceptance
Confidence versus prediction display: visual cues that separate proposed values from finalized labels in the review UI
Why D is correct
D is correct because the red-boxed feature in the referenced view corresponds to a suggested label, not a finished annotation. Suggestions are produced to accelerate labeling, especially for checkboxes, fields and tables, and remain provisional until a human validates them. Treating them as final would pollute training data, so the workflow is suggest then confirm, which matches vendor confirmation that these are label suggestions.
Why the others are wrong
A. Label annotation is wrong, because an annotation implies human confirmation, while the highlight is still a provisional proposal
B. Confidence level is wrong, because confidence is a numeric score or color intensity, not the suggested box itself
C. Label prediction is wrong in exam terminology, because predictions are runtime model outputs, while in labeling UI they surface as suggestions for training
UiSAIv1 exam tip
In labeling UI, suggestion equals proposal to accept, annotation equals human-confirmed truth.
8Which of the following are unstructured documents?
Invoices, receipts, purchase orders, and medical bills.
Banking forms, tax forms, surveys, and identity cards.
Contracts, emails, banking forms, and tax forms.
Contracts, agreements, and emails.
Answer: D
The short version
D — Contracts, agreements and emails are unstructured. They lack fixed field positions and vary freely in wording and layout, so they need ML understanding rather than fixed templates or form anchors.
Key concepts in this question
Unstructured documents: free-form content with no consistent layout or field coordinates across samples
Semi-structured examples: invoices, receipts, purchase orders and forms with predictable fields but variable layouts
Structured examples: tax forms, surveys and IDs with fixed boxes and positions suited to rules
Extraction strategy link: unstructured drives model-based extraction, structured favors rule or form extractors
Why D is correct
D is correct because contracts, agreements and emails have narrative paragraphs, clauses and varied formatting with no stable field anchors. UiPath docs list contracts, leases, annual reports, agreements and news as unstructured archetypes. Invoices and banking forms contain repeating totals, dates and boxes, so they fall into semi-structured or structured groups, not the fully unstructured set tested here.
Why the others are wrong
A. Invoices, receipts, purchase orders and medical bills are wrong, because those are classic semi-structured cases with detectable fields
B. Banking forms, tax forms, surveys and IDs are wrong, because fixed boxes make them structured, not free-form narrative
C. Mixed contracts plus banking and tax forms is wrong, because it blends unstructured with structured types instead of staying purely unstructured
UiSAIv1 exam tip
Narrative without boxes means unstructured, repeating fields means semi-structured, fixed boxes means structured.
9When creating a training dataset, what is the recommended number of samples for a Column field?
10-20 document samples per column field.
20-50 document samples per column field.
50-200 document samples per column field.
100-200 document samples per column field.
Answer: C
The short version
C — 50-200 samples per Column field. Column fields need far more examples than header or classification fields because table layout variation is high, and UiPath sets 50-200 per column as the planning baseline.
Key concepts in this question
Column fields: table columns such as item unit price and item quantity that repeat on every row and move with layout, so the extractor must see many positional variants.
Header fields: single occurrence fields such as invoice number or date that stabilize quickly and train with far fewer samples.
Layout diversity effect: clean simple layouts can succeed near 300 total samples for 5 columns, while complex diverse layouts can demand over 1000 documents.
Multilingual uplift: each added language needs roughly 200-300 extra samples covering all fields, so a 10-header plus 4-column model with 2 languages may need 600 to 1200 samples.
Why C is correct
C states the documented UiPath range of 50-200 samples per column field, matching vendor prose exactly. The range accounts for table variability: 50 covers clean uniform tables, 200 covers diverse ones, and totals scale with column count plus language coverage. Smaller ranges under-provide the row-level variation a table model needs, so C is the only planning figure that yields reliable extraction.
Why the others are wrong
A. 10-20 samples suits classification fields per class, not columns, and leaves table variance almost entirely unseen.
B. 20-50 samples suits regular header fields in simple cases, but is too thin for repeating column structures across layouts.
D. 100-200 samples narrows the low end incorrectly and would over-require simple cases, contradicting the documented 50-sample minimum per column.
UiSAIv1 exam tip
Remember columns as 50-200, classification as 10-20: columns cost the most data because every row and layout shift is a new training case.
10What additional property does the ML Extractor have compared to the other types of extractors?
Endpoint.
ML Skill.
Timeout.
ApiKey.
Answer: B
The short version
B — ML Extractor uniquely requires an ML Skill reference. Other extractors use endpoints, API keys or timeouts, but only the ML model path needs a deployed AI Center skill to call for predictions.
Key concepts in this question
ML Extractor activity: the Data Extraction Scope extractor that calls a deployed ML Skill endpoint
ML Skill property: the selector binding the extractor to a specific AI Center deployed model version
Form and Regex extractors: rule-based extractors configured with templates or patterns, not hosted models
Endpoint and ApiKey usage: generic connection settings that do not identify a specific skill deployment
Why B is correct
B is correct because Studio exposes ML Skill only on the Machine Learning Extractor to choose which deployed DocumentUnderstanding model to invoke. The skill encapsulates package version, deployment and URL, so the robot knows where to send the DOM and from where to receive extracted fields plus confidence. Form, Regex and Intelligent Form variants have no skill concept because their logic is local rules or built-in engines.
Why the others are wrong
A. Endpoint alone is wrong, because several activities use endpoints, so it is not distinctive to the ML Extractor
C. Timeout is wrong, because timeouts are common reliability settings across extractors and not model-specific
D. ApiKey is wrong, because keys authenticate calls generally and do not select the deployed ML model
UiSAIv1 exam tip
See ML Skill property and immediately choose Machine Learning Extractor.