A financial services firm is using UiPath Communications Mining to analyze customer emails. A model has been in production for six months with a 'Good' rating. Recently, the 'Balance' performance factor has dropped to 'Poor', while 'Performance' and 'Coverage' remain 'Good'. This is causing the model to over-predict common requests and miss newer, less frequent issues. What is the most appropriate first action in the 'Validation' page to address this specific problem?
Answer and explanation
Correct answer: B
The 'Balance' factor specifically measures how well the training data represents the overall dataset. A 'Poor' rating indicates some labels are over-represented while others are under-represented. The 'Rebalance' training mode is designed specifically to address this by surfacing communications from sparsely represented areas of the dataset, allowing the trainer to provide more examples for those labels and improve the model's balance.
Question 2
A developer is building a Document Understanding process using the DU Process Template. The requirement is to prevent any documents larger than 25 MB or with more than 100 pages from being processed to avoid excessive AI Unit consumption and potential timeouts. Where should this validation logic be implemented according to the template's best practices?
Answer and explanation
Correct answer: B
According to the best practices of the Document Understanding Process Template, initial validation checks on the input file (like size or page count) should be performed immediately after the transaction item (the document path) is retrieved. This happens in the Get Transaction Item state of the Main.xaml. Implementing the check here prevents the process from initiating the more resource-intensive DU workflows for invalid files, saving time and resources, and allowing the item to be correctly marked as a Business Rule Exception.
Question 3
Multiple answers
An automation architect needs to deploy a custom ML model for sentiment analysis, which was developed by an in-house data science team and packaged as a Docker container. The model is not part of the standard UiPath ML Packages. Which actions must be performed in AI Center to make this model available as an ML Skill? (Select TWO)
Answer and explanation
Correct answers: B, E
For custom models, the first step is to upload the code, dependencies, and the trained model itself as a single .zip file into the ML Packages section of AI Center. This makes the model's logic available within the platform.
After the ML Package is successfully uploaded, an ML Skill must be created. The ML Skill is the live, deployed version of the model that can be called by a UiPath Robot. This involves selecting the uploaded package and the specific version to deploy.
Question 4
True or False: In the UiPath Document Understanding Process template, the ShouldStop variable is exclusively used to gracefully stop the robot from Orchestrator and has no other function within the framework's logic.
Answer and explanation
Correct answer: B
False. While the ShouldStop variable is indeed used to handle the stop command from Orchestrator, it is also set to True within the framework's logic when no more transaction items are available in the queue or data source. This ensures the process ends correctly even when not manually stopped.
Question 5
A project requires extracting data from scanned, historical engineering diagrams. The text is often hand-written, uses non-standard fonts, and is sometimes skewed. The primary goal is to achieve the highest possible accuracy in text recognition before classification and extraction. Which OCR engine should be prioritized for this use case?
Answer and explanation
Correct answer: C
UiPath Document OCR is specifically designed for complex scenarios, including handwritten text, skewed documents, and varied fonts. It generally provides superior accuracy for scanned documents compared to other options like Tesseract or Microsoft OCR, making it the best choice for this challenging use case.
Question 6
A consultant is designing a Communications Mining taxonomy for an insurance company to analyze claims emails. The primary goal is to automate the initial routing and data entry for new claims. Which of the following represents the best-practice approach for designing the label hierarchy for this automation-focused use case?
Answer and explanation
Correct answer: D
For automation use cases, the best practice is to design a taxonomy where labels are mutually exclusive and directly map to a specific business process or automation workflow. This ensures that a single, clear trigger is identified for each communication, allowing a robot to confidently execute the correct downstream process (e.g., if the label is 'New Auto Claim', run the auto claim intake workflow). Granular details like damage type should be handled by extraction fields, not labels.
Question 7
Case Study: Global Logistics Document Automation
Company Background: Global Transport Inc. (GTI) is a multinational logistics company that processes thousands of shipping documents daily, including Bills of Lading (BOL), Commercial Invoices, and Packing Lists. These documents arrive from hundreds of different partners in various formats, ranging from high-quality structured PDFs to skewed, low-resolution scans. The data from these documents is manually entered into their Transportation Management System (TMS), leading to significant delays, data entry errors, and high operational costs.
Current Situation: GTI has initiated a project to automate this process using UiPath Document Understanding. The project team has successfully created a taxonomy and is now in the process of designing the core DU workflow. A key challenge is the high variability in document quality and layout. The same partner might send a clear, machine-readable invoice one day and a poorly scanned, handwritten BOL the next. The system must be resilient and require minimal human intervention for standard documents but also robustly handle exceptions.
Requirements:
The solution must classify each incoming document as either a BOL, Commercial Invoice, or Packing List.
For documents that cannot be classified with high confidence, a human user must be prompted for manual classification.
The solution must use a combination of rule-based and model-based extractors to maximize accuracy across different document layouts.
All extracted data must be validated against a set of business rules (e.g., 'Total Amount' must equal the sum of line items). Data that fails validation must be sent to a human for correction in Action Center.
The final, validated data must be exported as a JSON file for ingestion into the TMS.
Which workflow design most effectively meets all of GTI's requirements?
Answer and explanation
Correct answer: C
This option correctly addresses all requirements. 1) The Generative Classifier is ideal for classifying documents with high layout variability. 2) The Present Validation Station is the standard way to handle both low-confidence extraction and business rule failures, routing them to Action Center. 3) Combining a flexible ML Extractor with a precise Regex Based Extractor is a best practice for hybrid extraction scenarios. 4) The Export Extraction Results activity is the correct final step to generate the required JSON output.
Question 8
A developer is using the UiPath Communications Mining activities package in Studio. They need to process emails from a stream, but only those that have a predicted label of 'Urgent Inquiry' with a confidence score of 85% or higher. Which activity and property combination should be used to achieve this?
Answer and explanation
Correct answer: B
The correct and most efficient method is to define these criteria when the stream is created in the Communications Mining platform or via the Create Stream activity. The stream itself is configured to only include items that match the specified label ('Urgent Inquiry') and meet the minimum confidence threshold (0.85). The robot then simply polls this pre-filtered stream using Get Stream Items without needing to apply a client-side filter.
Question 9
During the 'Refine' phase of Communications Mining model training, an analysis of the 'Validation' page reveals the following metrics for a label named 'Address_Change':
Precision: 95%
Recall: 40%
F1 Score: 57%
What does this combination of metrics indicate about the model's performance for this label, and what is the recommended training action?
Answer and explanation
Correct answer: D
High precision (95%) means that when the model predicts 'Address_Change', it is almost always correct. Low recall (40%) means it fails to identify 60% of the actual 'Address_Change' communications. This indicates the model is too specific or cautious. The recommended action to improve recall is to use the 'Missed Label' training mode, which specifically helps you find examples that the model should have labeled but didn't.
Question 10
A developer has configured a Document Understanding solution where extracted data is sent to Action Center for validation. A business requirement is to add a note for the human validator, explaining a specific business rule for the 'Invoice_Total' field. How can this be achieved?
Answer and explanation
Correct answer: B
The Taxonomy Manager provides a specific feature called 'Validator notes' for each field. Any text entered into this property will be displayed as a helpful tooltip or instruction to the human user in the Action Center validation screen when they interact with that specific field. This is the designed method for providing context-specific guidance to validators.
