Task 1.3: Implement Data Validation and Processing Pipelines for FM Consumption. AIP-C01 Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A customer support company is building a chatbot using Amazon Bedrock. They have a large dataset of past support tickets in a raw, unstructured format. The tickets contain a lot of noise, such as typos, slang, and inconsistent formatting. They want to use this data to provide the chatbot with relevant context for answering new customer questions. What is the MOST effective way to enhance the quality of this historical data before using it for FM consumption?

   • AUse SageMaker Data Wrangler to create a visual flow that joins the ticket data with a customer database to add more context.
   • BUse Amazon Transcribe to convert the text tickets into audio and then back to text to correct any spelling errors.
   • CUse Amazon Comprehend to extract key entities like product names, error codes, and issue types, and then use a Bedrock FM to reformat the raw text into a clean, structured summary for each ticket.
   • DUse a Lambda function to remove all punctuation and convert all text to lowercase to create a normalized version of the tickets.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A research organization is building a pipeline to process a large corpus of scientific papers (PDFs) for a question-answering system using Amazon Bedrock. The pipeline extracts text from the PDFs and stores it in an S3 bucket. Before sending the text to the FM, they need to validate that the extracted text is of sufficient quality. They are concerned about pages that are scanned images with no extractable text, and pages with garbled text due to encoding issues. What is the most efficient way to detect these low-quality extractions?

   • AUse an AWS Glue Data Quality rule that checks the 'text_length' field for each page and flags pages where the length is below a certain threshold.
   • BUse Amazon Textract to re-analyze every page and compare the output with the original extraction to find discrepancies.
   • CUse a Lambda function to invoke a Bedrock FM to read the text and determine if it is garbled or nonsensical.
   • DUse Amazon Comprehend to detect the dominant language of the text and flag pages where the language is not English.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A company is developing a conversational AI assistant using Amazon Bedrock. They are using the Anthropic Claude 3 model. The application needs to maintain a conversation history and send it to the model for each new user message. The development team is unsure about the correct format for the conversation history in the API request. What is the correct way to format the messages?

   • AUse the 'conversation' parameter with a list of alternating 'human' and 'assistant' turns.
   • BUse the 'inputText' parameter and provide a JSON string that contains the full conversation history.
   • CUse the 'messages' parameter with an array of objects, each containing a 'role' ('user' or 'assistant') and 'content' field.
   • DSend the entire conversation history as a single string in the 'prompt' field, with each turn separated by a newline character.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   A developer wants to improve the consistency of text input to a foundation model by standardizing formatting, correcting spelling, and removing extra whitespace. The developer plans to use a serverless function to perform these normalization steps. Which AWS service can be used to run this custom normalization code without managing servers?

   • AAmazon SageMaker Processing
   • BAWS Lambda
   • CAWS Glue
   • DAmazon EC2
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A marketing company is using a SageMaker AI endpoint to host a custom text generation model. They have a pipeline that collects user-generated content (UGC) from social media. The UGC is often messy, containing emojis, URLs, and inconsistent casing. Before sending this data to the model for analysis, they want to validate and normalize it. They have set up a CloudWatch metric to track the number of records that fail validation. What is the most effective way to implement the validation and normalization logic?

   • AUse a Lambda function that is invoked for each new piece of UGC, cleans the data, and publishes a custom metric to CloudWatch for each validation check.
   • BUse Amazon Comprehend to detect the sentiment of the UGC and filter out any negative comments before sending them to the model.
   • CUse an AWS Glue Data Quality rule set to define the validation rules and run it as part of a Glue ETL job.
   • DCreate a SageMaker Processing job that runs on a schedule to clean the data, validate it, and publish the pass/fail metrics to CloudWatch.
   Reveal Solution Discussion Select an answer first