4. AI & ML Service

A healthcare company needs to process thousands of medical records daily. The records include handwritten notes, scanned PDF forms, and patient ID documents. The company wants an automated solution to extract text and structured data from these documents and store the results in Amazon S3 for downstream analytics. The solution must minimize the need to build custom ML models. Which solution meets these requirements?

• Use Amazon SageMaker Ground Truth to label the data and train a custom OCR model. Deploy the model using SageMaker endpoints.
• Use Amazon Textract to extract text, handwriting, and structured data directly from scanned documents and send the output to Amazon S3.
• Upload documents to an S3 bucket and configure an AWS Lambda function with Amazon Comprehend to extract text from the files.
• Use Amazon Rekognition to analyze scanned PDF forms and extract tables and text for storage in Amazon S3.

A media streaming company wants to automatically generate captions and transcripts for live and recorded video content. The solution must support multiple languages and redact personally identifiable information (PII) such as phone numbers from the transcripts. Which solution should the company use?

• Use Amazon Polly to generate speech from text and apply filters for PII removal.
• Use Amazon Comprehend to detect entities and redact PII before storing the video content in Amazon S3.
• Use Amazon Transcribe with automatic language identification and PII redaction to create transcripts and captions.
• Use Amazon Rekognition to detect objects in video frames and produce text transcripts.

A retail company wants to build a customer service chatbot that can provide accurate answers using the company’s product manuals and FAQs. The company does not want to train a custom model and prefers a managed service that supports Retrieval-Augmented Generation (RAG) to pull real-time information from Amazon S3. Which solution should the company use?

• Use Amazon Lex integrated with a Lambda function that queries the product manuals.
• Use Amazon SageMaker to fine-tune an open-source foundation model with the company’s product data.
• Use Amazon Bedrock with a knowledge base to enable RAG and connect to the product manuals stored in S3.
• Use Amazon Translate to convert documents into multiple languages for the chatbot.

A financial services company wants to customize a foundation model to generate reports in a specific writing style aligned with its brand. The company requires fine-tuning the model with its internal data while keeping control of data privacy. Which approach using Amazon Bedrock meets these requirements?

• Use Amazon Comprehend to classify financial text and generate branded reports.
• Use Amazon Bedrock to fine-tune a foundation model with company data stored in S3 using Provisioned Throughput.
• Use Amazon SageMaker Autopilot to train a new model from scratch with the financial data.
• Use Amazon Translate to adapt reports into different languages with a customized style.

An e-commerce company wants to automate order management tasks such as retrieving recent purchases, placing new orders, and checking return policies. The solution must coordinate multi-step tasks, integrate with APIs, and use knowledge bases when needed. Which AWS service should the company use?

• Use Amazon Lex integrated with AWS Lambda functions for task automation.
• Use Amazon Bedrock Agents configured with predefined action groups and knowledge bases.
• Use Amazon Comprehend to extract intents from customer orders.
• Use Amazon SageMaker to train a custom orchestration model.

A global education platform wants to ensure that its AI-powered chatbot built on Amazon Bedrock does not provide harmful content or share personally identifiable information (PII). The solution must block restricted topics and reduce hallucinations. Which feature should the company implement?

• Use Amazon Translate to filter out unwanted content.
• Apply Amazon Bedrock Guardrails to control model interactions and remove PII.
• Use Amazon Kendra to restrict the search scope of the chatbot.
• Fine-tune the model with labeled data for safe responses.

A startup is experimenting with different foundation models on Amazon Bedrock for unpredictable workloads. The team wants to avoid long-term commitments and pay only for actual usage. Which pricing option should the team choose?

• Batch prediction with up to 50% discount
• Provisioned Throughput with reserved capacity
• On-Demand pricing with pay-per-use
• Purchase EC2 Reserved Instances for Bedrock

A company wants to assess the quality of text summarization generated by its foundation model in Amazon Bedrock. The solution must use built-in evaluation methods and provide automatic scoring with metrics such as ROUGE and BLEU. Which feature should the company use?

• Bedrock Guardrails
• Bedrock Model Evaluation
• Bedrock Provisioned Throughput
• Amazon Comprehend Custom Classification

A multinational bank receives millions of chat messages and support emails every month in different languages. The compliance team needs to automatically detect sensitive financial terms (e.g., account numbers, policy IDs), classify the messages by category (fraud, loan, inquiry), and perform sentiment analysis in near real time. The solution must minimize operational overhead and avoid building ML models from scratch. Which solution should the company implement?

• Use Amazon Translate to normalize all text into English and run a custom classification model on Amazon SageMaker.
• Use Amazon Textract to extract entities from all emails and Amazon Kendra to classify them by intent.
• Use Amazon Comprehend with built-in features for entity recognition, custom classification, and sentiment analysis across multiple languages.
• Store all text in Amazon S3 and run Athena queries with SQL functions for keyword matching and sentiment scoring.

A global video-sharing platform needs to automatically moderate millions of daily uploads. The solution must detect violent or adult content in videos, flag potential violations for human review, and integrate with a custom workflow for regulatory compliance. The company also requires face recognition to block banned users and celebrity recognition for tagging. Which AWS service combination best meets these requirements with the least custom development?

• Use Amazon SageMaker to train a custom vision model for moderation and celebrity recognition.
• Use Amazon Rekognition Content Moderation for unsafe content detection, face recognition for banned users, and celebrity recognition APIs.
• Use AWS Panorama to run video analytics on-premises and push results to Amazon CloudWatch.
• Use Amazon Transcribe to convert audio tracks to text and Amazon Comprehend to detect sensitive content.