Skill 2.1.8: Apply AWS AI services to solve specific business problems (for example, Amazon Textract, Amazon Rekognition, Amazon Comprehend, Amazon Transcribe)
| Question | Answer & Explanation |
|---|---|
| ### Question 1 - Document Extraction with Least Operational Overhead A financial technology company receives thousands of scanned loan applications in PDF format daily. They need to automatically extract key-value pairs like applicant name, loan amount, and interest rate, as well as table data from these documents. The engineering team has strict constraints to build this pipeline with the least operational overhead and fastest time-to-market. Which AWS service should they use to accomplish this? - A) Train a custom computer vision model using Amazon SageMaker to detect and extract text and table structures. - B) Use Amazon Textract's AnalyzeDocument API to extract text, tables, and key-value pairs.- C) Use Amazon Rekognition's DetectText API to extract text from the scanned PDFs.- D) Fine-tune a foundation model in Amazon Bedrock by providing sample PDFs and expected JSON outputs. |
Correct Answer: B Correct Answer Analysis: - B is Correct: Amazon Textract is a fully managed service specifically designed to automatically extract text, handwriting, layout elements, and structured data (like tables and key-value pairs) from scanned documents. It requires no model training, providing the least operational overhead and fastest time-to-market. Distractor Analysis: - A is Incorrect: Training a custom CV model in SageMaker requires data labeling (bounding boxes), model architecture selection, training, and hosting infrastructure, which introduces massive operational overhead compared to a managed API. - C is Incorrect: While Amazon Rekognition DetectText can extract raw text from images, it is designed for short text in natural scenes (like street signs or social media images), not for extracting structured tables and key-value pairs from complex forms.- D is Incorrect: Foundation models in Bedrock are primarily for text generation/reasoning. While some support multimodal inputs, fine-tuning them for precise structured document extraction is unnecessarily complex, costly, and inefficient compared to Textract. > Key Takeaway: For extracting structured data (tables, forms, key-value pairs) from scanned documents, Amazon Textract is the native, low-overhead AWS AI service. |
| ### Question 2 - Brand Logo Detection in Video Streams A marketing agency wants to analyze user-generated video content to measure the frequency and duration of specific brand logos appearing in the footage. The agency needs a solution that allows them to train a custom detector for their proprietary brand logos without managing the underlying infrastructure. The constraint is minimal ML expertise required for setup. Which AWS service is the MOST appropriate to build this custom logo detection model? - A) Amazon Rekognition Custom Labels - B) Amazon Comprehend Custom Classification - C) Amazon Textract AnalyzeDocument - D) Amazon SageMaker Ground Truth with a custom PyTorch training script (Select ONE) |
Correct Answer: A Correct Answer Analysis: - A is Correct: Amazon Rekognition Custom Labels allows users to train a custom model to identify specific objects, scenes, and concepts (like proprietary brand logos) in images and videos. It abstracts away the model training and hosting infrastructure, requiring minimal ML expertise. Distractor Analysis: - B is Incorrect: Amazon Comprehend is a natural language processing (NLP) service. It cannot analyze visual data like video footage or images for logo detection. - C is Incorrect: Amazon Textract is designed for extracting text and structured data from documents, not for identifying specific visual objects or brand logos in video streams. - D is Incorrect: Building a custom training script in SageMaker requires deep ML engineering expertise to architect, train, and deploy a computer vision model, violating the "minimal ML expertise" and "no infrastructure management" constraints. > Key Takeaway: When you need to detect custom, proprietary objects in images/videos without managing infrastructure, use Amazon Rekognition Custom Labels. |
| ### Question 3 - Real-Time Call Transcription and PII Redaction A healthcare company operates a call center and needs to transcribe patient-agent phone calls in real-time. Compliance regulations mandate that Protected Health Information (PHI) and Personally Identifiable Information (PII) must be redacted directly in the transcript. The solution must be a fully managed AWS AI service that natively understands medical terminology. Which service configuration meets these requirements? - A) Amazon Transcribe Medical with Content Redaction enabled - B) Amazon Transcribe Standard with a custom SageMaker NLP post-processing pipeline - C) Amazon Comprehend Medical to process the audio directly - D) Amazon Bedrock to generate medical summaries from audio files (Select ONE) |
Correct Answer: A Correct Answer Analysis: - A is Correct: Amazon Transcribe Medical is specifically trained to understand medical terminology. It supports real-time transcription and has a native Content Redaction feature that automatically identifies and redacts PHI/PII directly in the generated text, satisfying all constraints as a fully managed service. Distractor Analysis: - B is Incorrect: Transcribe Standard does not natively understand medical terminology as accurately as Transcribe Medical, and routing through a custom SageMaker NLP pipeline introduces unnecessary operational overhead and latency, violating the fully managed constraint. - C is Incorrect: Amazon Comprehend Medical is an NLP service that analyzes text, not audio. It cannot directly transcribe audio files. - D is Incorrect: Bedrock foundation models process text and images (via multimodal models), not direct audio streams. You would still need a transcription service first. > Key Takeaway: For audio transcription involving medical terminology and requiring native PII/PHI redaction, Amazon Transcribe Medical is the correct managed service. |
| ### Question 4 - Sentiment Analysis on Customer Feedback An e-commerce company receives thousands of free-text product reviews daily. They need to automatically categorize the sentiment of these reviews (positive, negative, neutral) and extract the main product features mentioned (e.g., "battery life", "screen resolution"). The architecture must be cost-effective and avoid the overhead of training custom models. Which AWS service provides these out-of-the-box capabilities? - A) Amazon Rekognition - B) Amazon Comprehend - C) Amazon Textract - D) Amazon SageMaker JumpStart (Select ONE) |
Correct Answer: B Correct Answer Analysis: - B is Correct: Amazon Comprehend is a managed natural language processing (NLP) service that provides out-of-the-box APIs for sentiment analysis, key phrase extraction, and entity recognition. It requires no custom model training, making it highly cost-effective and low-overhead. Distractor Analysis: - A is Incorrect: Amazon Rekognition is for image and video analysis, not text processing. - C is Incorrect: Amazon Textract is for extracting text from scanned documents, not for analyzing the sentiment or meaning of text data. - D is Incorrect: While SageMaker JumpStart offers pre-trained models, deploying and managing the endpoint still requires significantly more operational overhead than calling a fully managed serverless API like Amazon Comprehend. > Key Takeaway: For out-of-the-box text NLP tasks (sentiment, entities, key phrases), use Amazon Comprehend to avoid training and hosting overhead. |
| ### Question 5 - Automated Document Processing Pipeline (Select TWO) An insurance company needs to automate the processing of incoming claim forms. The forms arrive as scanned PDFs. The architecture must extract the raw text and form fields from the PDF, and then analyze that extracted text to identify specific medical conditions and dosages mentioned in the claim. Which TWO AWS services should be combined to build this pipeline with the least operational overhead? - A) Amazon Textract - B) Amazon Comprehend Medical - C) Amazon SageMaker Training Job - D) Amazon Rekognition Custom Labels - E) Amazon Transcribe (Select TWO) |
Correct Answer: A and B Correct Answer Analysis: - A is Correct: Amazon Textract is needed first to extract the raw text and key-value pairs from the scanned PDF document. - B is Correct: Amazon Comprehend Medical is then used to analyze the extracted text to identify medical conditions, medications, and dosages using its pre-trained medical NLP models. Distractor Analysis: - C is Incorrect: A SageMaker Training Job requires model development and training overhead, which violates the "least operational overhead" constraint. - D is Incorrect: Rekognition Custom Labels is for visual object detection in images, not for text extraction or medical text analysis. - E is Incorrect: Amazon Transcribe is for speech-to-text conversion, not for processing scanned PDF documents. > Key Takeaway: Document AI workflows often chain Textract (to get text out of PDFs) with Comprehend/Comprehend Medical (to extract entities and meaning from the text). |
| ### Question 6 - Image Content Moderation A social media startup is building an automated pipeline to filter user-uploaded images. The pipeline must detect explicit, violent, or inappropriate content before the images are published to the public feed. The startup has a strict requirement for a fully managed, serverless API that can easily scale and requires no custom model training. Which AWS service is the best fit for this requirement? - A) Amazon Comprehend - B) Amazon Rekognition Image Moderation - C) Amazon Textract - D) Amazon SageMaker JumpStart (Select ONE) |
Correct Answer: B Correct Answer Analysis: - B is Correct: Amazon Rekognition provides a Content Moderation API that is pre-trained to detect explicit, suggestive, violent, and other inappropriate content in images. It is fully managed, serverless, and requires no custom training. Distractor Analysis: - A is Incorrect: Amazon Comprehend is an NLP service for analyzing text, not images. - C is Incorrect: Amazon Textract extracts text from documents; it does not perform visual content moderation or detect inappropriate imagery. - D is Incorrect: While SageMaker JumpStart has pre-trained models, deploying them requires managing endpoints and infrastructure, whereas Rekognition is a simple API call with zero infrastructure management. > Key Takeaway: For automated image safety and content moderation, the pre-trained Amazon Rekognition Content Moderation API is the standard managed solution. |
| ### Question 7 - Extracting Custom Entities from Text A manufacturing company wants to automatically scan internal engineering reports (stored as text) to extract proprietary part numbers (e.g., "XJ-9920-A") and warranty codes. Standard entity recognition services do not recognize these custom formats. The company has a dataset of 1,000 annotated reports but lacks a dedicated ML engineering team. The solution must use a managed AWS AI service to train this custom model. Which approach satisfies these requirements? - A) Use Amazon Comprehend Custom Entity Recognition to train a model on the annotated reports. - B) Use Amazon SageMaker to train a custom SpaCy NLP model and deploy it on an endpoint. - C) Use Amazon Textract to search for regular expressions matching "XJ-XXXX-X". - D) Use Amazon Bedrock to fine-tune a foundation model on the engineering reports. (Select ONE) |
Correct Answer: A Correct Answer Analysis: - A is Correct: Amazon Comprehend Custom Entity Recognition allows users to train a custom model to extract specific, proprietary entities (like internal part numbers) using their own annotated documents, without needing ML expertise or managing infrastructure. Distractor Analysis: - B is Incorrect: Using SageMaker to train and deploy a custom SpaCy model requires deep ML engineering expertise and infrastructure management, violating the constraints. - C is Incorrect: Textract is for extracting text from documents, not for running regex or NLP entity extraction. Even if text is extracted, regex is brittle and doesn't constitute a managed AI training service for entities. - D is Incorrect: Fine-tuning a Bedrock foundation model for simple entity extraction is overkill, highly expensive, and less reliable than a dedicated NLP custom entity recognition service. > Key Takeaway: When pre-trained NLP models don't recognize your proprietary entities, use Amazon Comprehend Custom Entity Recognition for a managed training experience. |
| ### Question 8 - Call Center Analytics and Summarization A telecom company wants to extract insights from customer service calls. They need a solution that provides automated transcription, speaker identification (agent vs. customer), sentiment analysis, and call summarization. The key constraint is lowest operational overhead, requiring a native AWS service purpose-built for this workflow rather than a custom pipeline. Which AWS service provides these specific built-in features? - A) Amazon Transcribe Call Analytics - B) Amazon Comprehend - C) Amazon Rekognition Video - D) Amazon Polly (Select ONE) |
Correct Answer: A Correct Answer Analysis: - A is Correct: Amazon Transcribe Call Analytics is a specialized feature of Amazon Transcribe designed specifically for call centers. It natively provides transcription, speaker diarization (partitioning), sentiment analysis, call summarization, and characterizes call drivers in a single managed API. Distractor Analysis: - B is Incorrect: Amazon Comprehend performs sentiment analysis and key phrase extraction, but it does not perform audio transcription, speaker identification, or call summarization out-of-the-box. Combining it with Transcribe would create a custom pipeline, violating the lowest overhead constraint. - C is Incorrect: Rekognition is for computer vision (images/video), not audio or call center analytics. - D is Incorrect: Amazon Polly is a Text-to-Speech service (the opposite of what is needed here). > Key Takeaway: Amazon Transcribe Call Analytics is the purpose-built AWS service for comprehensive call center audio analysis (transcription, sentiment, summarization). |
| ### Question 9 - Video People Identification A corporate security team needs to monitor security camera footage to identify employees as they enter a restricted facility. They have a collection of employee face images. They need a solution that can index these faces and search the video streams in near real-time with minimal infrastructure management. Which AWS service should they use to implement this face search capability? - A) Amazon Rekognition Face Search - B) Amazon Textract - C) Amazon SageMaker with a pre-trained ResNet model - D) Amazon Comprehend Medical (Select ONE) |
Correct Answer: A Correct Answer Analysis: - A is Correct: Amazon Rekognition Face Search allows you to create a face collection (index) of known individuals and search incoming images or video streams against this collection. It is fully managed and handles the underlying vector search and face matching natively. Distractor Analysis: - B is Incorrect: Amazon Textract extracts text and forms from documents; it does not perform facial recognition or biometric search. - C is Incorrect: Deploying ResNet on SageMaker for facial recognition requires building a custom vector database, embedding pipeline, and hosting infrastructure, which violates the minimal infrastructure management constraint. - D is Incorrect: Amazon Comprehend Medical is for analyzing medical text, completely unrelated to computer vision or video processing. > Key Takeaway: For indexing and searching faces in images or video streams, use Amazon Rekognition Face Search to avoid managing custom vector databases. |
| ### Question 10 - Multi-modal Meeting Processing (Select TWO) A global enterprise needs to process recorded video files of internal virtual meetings. The required outputs are a searchable text transcript of the conversation, and the identification of which specific employee was speaking during each segment. The company wants to use AWS AI services with the least operational overhead and avoid custom ML development. Which TWO AWS services should they combine to achieve this? - A) Amazon Transcribe - B) Amazon Rekognition Video - C) Amazon SageMaker Ground Truth - D) Amazon Polly - E) Amazon Comprehend Custom Classification (Select TWO) |
Correct Answer: A and B Correct Answer Analysis: - A is Correct: Amazon Transcribe is used to convert the audio track of the recorded meeting video into a searchable text transcript. - B is Correct: Amazon Rekognition Video can process the video stream to identify faces (using a pre-indexed face collection of employees), providing the identity of the person speaking on camera during different segments. Distractor Analysis: - C is Incorrect: SageMaker Ground Truth is a data labeling service, not an inference or processing service. It requires human annotators and does not automatically identify faces or transcribe audio. - D is Incorrect: Amazon Polly is a Text-to-Speech service (generating audio from text), which is the reverse of what is needed. - E is Incorrect: Amazon Comprehend Custom Classification classifies text into custom categories; it does not perform audio transcription or visual facial recognition. > Key Takeaway: When processing meeting videos, pair Amazon Transcribe (for audio-to-text) with Amazon Rekognition Video (for visual speaker identification) to build a low-overhead multi-modal pipeline. |