Option B provides the strongest match among the listed alternatives because it combines managed speaker identification, multimodal generative analysis, and a native search-oriented datastore without unnecessary custom orchestration.

Amazon Transcribe speaker diarization distinguishes speakers and assigns identifiers such as spk_0 , spk_1 , and so forth. Its output includes speaker labels and timestamps, allowing individual statements to be associated with particular speakers and positions in the meeting timeline.

Anthropic Claude models on Amazon Bedrock support multimodal prompts that combine text and images. The application can therefore supply speaker-attributed transcript segments together with extracted presentation/video frames and instruct the model to generate summaries that associate spoken comments with relevant visual information. AWS ' s Claude Messages API documentation explicitly supports mixed image-and-text input.

Amazon OpenSearch Service is appropriate for storing the enriched meeting records and supporting document indexing and full-text searching. It provides substantially more natural full-text retrieval than attempting to create a custom indexing layer over DynamoDB.

A uses Transcribe and Rekognition effectively, but it states that a Lambda function itself “generates” the summaries without specifying a generative model and requires custom correlation logic. C chains BDA, Transcribe, Rekognition, DynamoDB, and Bedrock, increasing service count and operational complexity; DynamoDB also does not natively satisfy the full-text-search requirement. D similarly depends on DynamoDB plus a custom indexing mechanism.

Current AWS capabilities make Bedrock Data Automation even more capable than the wording of C suggests: BDA can produce video summaries, scene-level summaries, detected video text, full audio transcripts, and speaker identification. However, among the supplied choices , B remains the cleanest architecture satisfying multimodal GenAI analysis, diarization, scalable managed processing, and native searchable storage.

==========