AI processing
Turn documents and recordings into structured feedback insights.
The analysis pipeline prepares uploaded content before asking a foundation model to interpret it.
Prepare the input
- Documents: Textract extracts text and document structure from supported files.
- Audio: Transcribe converts supported audio into text.
- Video: prepare the audio in a supported format for transcription; video handling may need a separate preprocessing step.
The appropriate path depends on the uploaded format. Keep the original file in S3 and track each processing stage so a failed extraction does not appear as a completed analysis.
Analyze with Bedrock
Amazon Bedrock provides access to foundation models from Amazon and other providers, including Anthropic. Model capabilities and availability vary. The workshop uses this layer to turn extracted text or transcripts into structured feedback insights. See the Bedrock overview.
The model can summarize feedback or identify recurring themes, but its output still needs validation before storage or display. Choose a response structure that the application can consume and retain enough context for organizers to understand the result.
Store results and notify
The worker stores processed metadata and insights in DynamoDB, associated with the original file and event. Only after successful persistence does it publish a completion notification to SNS.
Keep extraction, transcription, and model analysis as identifiable stages. Some service operations are asynchronous, so the workflow must track completion rather than assume every operation finishes during one Lambda invocation.