Vertex Macro｜Amarathon 2025 Recap 02：Amazon Bedrock Data Automation

Problem Statement

● Organisations struggle with unstructured data in various formats (documents, images, audio, video).

● Manual processing is slow, inconsistent, and costly.

● Existing automation systems are rigid, requiring templates, rules, and manual corrections.

● Increasing demand for compliance, accuracy, and scalability.

● Need for automating multi-format data processing with high accuracy using generative AI.

What is Bedrock Data Automation (BDA)?

● A fully-managed document and media automation capability in Amazon Web Services.

● Enables building end-to-end extraction, classification, and transformation pipelines using foundation models.

● Processes documents, images, audio, and video at scale.

● Orchestrates multi-step workflows using serverless automation.

● Minimises custom code while maximising flexibility.

Input Asset

[ 1 ] Supports various formats:

● Documents (PDF, DOCX, scanned, structured/unstructured)

● Images (PNG, JPG)

● Audio (voice notes, call recordings)

● Video (meetings, CCTV, webinars)

[ 2 ] Offers two types of output instructions:

● Standard Output Configuration

● Custom Schema based on matched blueprint

Output Response

● Linearized Text representation of the asset based on configuration.

● Output returned as JSON + additional files if selected in configuration.

● Supported Formats & Information BDA Extracts:

[ 1 ] Documents:

● Extracts fields, tables, entities

● Classifies, transforms, summarises, and validates

[ 2 ] Images:

● Offers OCR, document classification, object detection, and handwriting extraction

[ 3 ] Audio:

● Provides transcription, summarisation, sentiment analysis, speaker detection, and intent extraction

[ 4 ] Video:

● Offers video summaries, speech-to-text, scene detection, object recognition, and action understanding

Standard Output vs Custom Output (Blueprints)

[ 1 ] Standard Output:

● Out-of-the-box extraction

● Ideal for common documents

● Zero setup, quick results

[ 2 ] Custom Output:

● Based on blueprints

● Allows for prompt or user-defined blueprints

● Accelerates setup and maintains consistency

● Suitable for industry-specific or complex documents

---

Types of Document Blueprints

[ 1 ] Classification:

● Invoice, bank statement, ID card, contract, HR letter, etc.

[ 2 ] Extraction:

● Entities, fields, tables, metadata

[ 3 ] Transformation:

● Modify or restructure data

[ 4 ] Normalization:

● Standardise data values

[ 5 ] Validation:

● Validate extracted fields against rules

Use Cases

[ 1 ] Banking & Finance:

● Automate bank statements, invoices, receipts, fraud checks

[ 2 ] Insurance:

● Claims processing from forms, photos, reports

● Auto-summaries, extraction, validation

[ 3 ] Customer Support:

● Transcribe & summarize calls

● Detect sentiment and customer intent

[ 4 ] HR & Legal:

● Process resumes, contracts, offer letters

● Extract skills, clauses, obligations

[ 5 ] Security & Operations:

● Summaries from meeting recordings

● CCTV context extraction (people, actions)

Key Takeaways

● Bedrock Data Automation (BDA) is a comprehensive, customizable, and scalable solution.

[ 1 ] One Platform for All Formats:

● Automates document, image, audio, and video processing.

[ 2 ] Customizability:

● Delivers highly accurate and customizable outputs using advanced foundation models.

● Ensures trustworthy and consistent insights tailored to any business workflow.

[ 3 ] Enterprise-Ready:

● Scales to thousands of files with high accuracy and compliance.

[ 4 ] Faster, Cheaper, Smarter:

● Reduces manual workload and delivers clean, structured outputs instantly.