Task 1.3: Implement Data Validation and Processing Pipelines for FM Consumption. AIP-C01 Practice Questions (Page 5)

Questions 21–21

1. 21expert · hard

   A media company is building a search application for a large archive of video content. The application must support searching by natural language queries (e.g., 'a red car driving on a highway'). The pipeline needs to process video files to create a searchable index. The team is evaluating two approaches: (1) using Amazon Rekognition Video to detect labels and Amazon Transcribe to generate a transcript, then embedding the combined text; (2) using Amazon Bedrock to analyze key frames and generate a text description for each frame. The team has a limited budget and needs to process a large volume of video. Which approach is more cost-effective and why?

   • AApproach 1 is more cost-effective, but it will not work because Rekognition cannot process video files directly.
   • BApproach 2, because using a powerful FM like Claude will generate much more accurate and detailed descriptions than Rekognition's labels, leading to better search results.
   • CBoth approaches are equally cost-effective, as the cost is primarily determined by the amount of video data being processed, not the specific services used.
   • DApproach 1, because Rekognition and Transcribe are purpose-built for video and audio analysis and are generally more cost-effective for large volumes than invoking a generative FM for every frame.
   Reveal Solution Discussion Select an answer first