Exams / AIP-C01 / Study guide / Processing images, audio, documents and tables for FMs (Bedrock Data Automation, Transcribe)
Processing images, audio, documents and tables for FMs (Bedrock Data Automation, Transcribe): AIP-C01 study notes and practice questions
What to know about Processing images, audio, documents and tables for FMs (Bedrock Data Automation, Transcribe) for the AIP-C01 exam, from the Foundation Model Integration, Data Management, and Compliance domain (31% of the exam guide), then 1 free question to test it. Part of the AIP-C01 study guide.
AIP-C01 gives an input type and a downstream goal, such as searchable call recordings, answers about an image, or field extraction from forms. Match the processing tool to the input and to the output shape you need.
Choose the processing tool
Bedrock Data Automation (BDA) extracts structured output from documents, images, audio and video using blueprints you define. Choose it for consistent structured results with little custom code.
Amazon Transcribe converts speech to text, with options such as speaker diarization and PII redaction. Send the transcript to a foundation model to summarize or answer questions.
Amazon Textract extracts text, forms (key-value pairs) and tables from scanned documents. Use AnalyzeDocument with the FORMS and TABLES features for structured fields.
Multimodal foundation models accept image and text in one request, for example a question about a chart. They suit understanding, not high-volume exact extraction.
Foundation model parser for a Bedrock knowledge base reads complex tables and figures in PDFs during ingestion. Use it when standard parsing loses table structure.
Image questions
Send the image as an image content block with the question in the same Converse request. Check the model’s supported formats and size limits. For video, either sample frames for a multimodal model or use Bedrock Data Automation.
Scale with Step Functions
For many media files: an S3 upload event starts a Step Functions workflow. A Map state processes each file in parallel, calls BDA or Transcribe as asynchronous jobs that the workflow waits on, then writes results to S3 and indexes them. Retry and Catch handle transient failures, and failed items go to a dead-letter path for review.
How to choose
Scenario wording
Choose
“Extract fields from invoices and video into one JSON schema”
Bedrock Data Automation
“Summarize recorded support calls”
Transcribe, then a foundation model for the summary
“Key-value pairs from scanned forms”
Textract AnalyzeDocument with FORMS
“Answer questions about a photo”
Multimodal model with image input
“Knowledge base PDFs with complex tables”
Foundation model parser
“Thousands of hours of media every night”
Step Functions Map state with asynchronous jobs
Common traps
Sending raw audio to a text-only model. Transcribe first.
Using Textract to reason about meaning. It extracts; a model interprets.
Using a multimodal model for bulk exact extraction where BDA or Textract is cheaper and more consistent.
Ignoring file size and page limits, which differ by service.
Calling long media synchronously. Use asynchronous jobs and wait for completion.
A property insurer receives three kinds of claim material: scanned PDF claim forms with tables, photos of storm damage, and recorded phone calls with policyholders.
The claims system needs the same core fields from all three, such as policy number, incident date, amount claimed and damage category, in one JSON format that downstream systems can read. The team has only two developers and wants to write as little code as possible.
New file types are rare, but the fields must be extracted from each file with a consistent schema.
Which solution meets these requirements with the LEAST custom code?
Choose one answer.
AA. Use Amazon Bedrock Data Automation with a blueprint for each material type, so that documents, images and audio each return structured JSON with the same core fields in S3.
BB. Run Amazon Textract on the PDF forms, Amazon Transcribe on the call recordings, and a custom computer vision model on SageMaker AI for the photos, then join the three outputs with an AWS Glue job.
CC. Send each file as raw bytes in a single prompt to a foundation model, and ask it to return the policy number, date, amount and damage category as JSON for every claim.
DD. Convert the PDFs to plain text with a script, transcribe the calls with Amazon Transcribe, and skip the photos, because the core claim fields can be filled from the forms and the calls.
Show the answer and explanation
Answer: A
A. Right: Bedrock Data Automation extracts structured output from documents, images, audio and video using blueprints, so the team writes configuration rather than extraction code. The consistent JSON output meets the schema and effort requirements. Learn more: Amazon Bedrock Data Automation (opens in a new tab)
B. Wrong: This approach can work, but it needs three extraction pipelines, a trained image model and a join job to maintain. That is far more code than the two developers can support, so it misses the effort requirement.
C. Wrong: Large audio files and multi-page scans may exceed what one request can carry, and output consistency would depend on the prompt. The team would have no reliable schema check, so the approach misses the consistency requirement.
D. Wrong: This option is simpler, but photos often show the damage category that the claims system needs. Dropping them means the required fields are incomplete for many claims, which breaks the schema requirement.
Explanation
Bedrock Data Automation handles documents, images and audio with blueprints that output consistent JSON, so the team configures rather than codes the extraction. The multi-service pipeline and the raw prompt approach each add code or risk, and skipping photos leaves required fields empty.
Want more on Processing images, audio, documents and tables for FMs (Bedrock Data Automation, Transcribe)? The 5 full-length practice sets have 6 more Processing images, audio, documents and tables for FMs (Bedrock Data Automation, Transcribe) questions, 375 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →
Study Processing images, audio, documents and tables for FMs (Bedrock Data Automation, Transcribe)