SERVERLESS AI · DOCUMENT AUTOMATION

AI Document Processing Pipeline

Event-driven AWS pipeline that extracts document text with Textract, orchestrates processing with Step Functions, classifies and summarizes content with Amazon Bedrock, and writes normalized JSON results to protected S3 storage.

Amazon S3EventBridgeStep FunctionsTextractLambdaAmazon BedrockCloudWatchTerraform

View AWS Architecture ↓

PROJECT STATUSImplemented

CI VALIDATIONLoading…

DEPLOYMENTManual by Design

OUTPUTStructured JSON

PROCESSING FLOW

Upload to structured AI output

INGESTS3 Input

Encrypted, versioned document storage.

ROUTEEventBridge

Object-created events launch processing.

EXTRACTTextract

Asynchronous text extraction for PDFs/images.

ORCHESTRATEStep Functions

Starts, waits, polls, and handles failures.

ENRICHLambda + Bedrock

Classification, summary, actions, confidence.

STORES3 Output

Normalized JSON under processed/.

AI Document Processing Pipeline data flow showing S3 input, EventBridge, Step Functions, Amazon Textract, AWS Lambda, Amazon Bedrock, normalized JSON, S3 output, failure handling, and CloudWatch observability

AWS architecture diagram for AI Document Processing Pipeline showing S3 input, EventBridge, Step Functions, Amazon Textract, AWS Lambda, Amazon Bedrock, S3 output, CloudWatch, IAM, Terraform, and GitHub Actions

Implemented controls

Protected StorageEncryption, versioning, Block Public Access.

Least Privilege IAMSeparate EventBridge, Step Functions, and Lambda roles.

Grounded PromptingBedrock uses only Textract-derived source text.

Output ValidationLambda normalizes model JSON before storage.

Failure HandlingFailed Textract jobs stop before AI processing.

CI ValidationUnit tests, Bandit, Terraform fmt/init/validate.

Implementation and validation status are shown here; the stack is not claimed as production-deployed.

Recent validation runs

Run	Result
Loading…

Reading public GitHub Actions metadata.

MICROSOFT AZURE REFERENCE ARCHITECTURE

Azure reference equivalent for portfolio comparison. The original project implementation remains as documented elsewhere on this page.