Week 12 Worklog - DocuFlow AI Testing, Final Demo, and Wrap-up

Week 12 Objectives:

• Finalize the last parts of DocuFlow AI before the demo: workshop guide, evidence, test cases, and final runbook.
• Run end-to-end tests for the main flows: login, upload, queue, workflow, Textract, AI Proxy, result storage, review, alert, and cleanup.
• Prepare the final report, 7-10 minute demo script, and resource cleanup checklist to control cost.

Tasks to be carried out this week:

Week 12 Achievements:

• Finalized the DocuFlow AI workshop content:

  • Updated the overview according to the admin-approved architecture.
  • Synchronized prerequisites, deploy steps, AI Proxy setup, observability, and cleanup.
  • Rechecked internal links, architecture images, sample commands, and expected results.
  • Kept Vietnamese and English content consistent.

• Tested the end-to-end happy path:

  • The user logs in with Cognito and accesses the protected frontend.
  • The frontend calls POST /documents/upload-url, receives a presigned URL, and uploads an invoice to S3 Raw.
  • The S3 ObjectCreated event flows through EventBridge, SQS, and Job Starter Lambda.
  • Step Functions runs validation, Textract AnalyzeExpense, AI Proxy Lambda, External AI API, JSON validation, confidence/status, and result persistence.
  • DynamoDB stores metadata/status EXTRACTED; S3 Processed stores result.json using the agreed path convention.

• Tested failure path and review path:

  • Invalid file type or invalid metadata is rejected or moved to FAILED.
  • A low-quality or missing-field receipt moves to REVIEW_REQUIRED with reviewReasons.
  • External AI timeout/rate limit/invalid JSON is handled by workflow retry/catch and clear errorType logs.
  • The reviewer can correct fields and move the result to CORRECTED or APPROVED.

• Collected report evidence:

  • Frontend: login, upload, document list, document detail, review page, and error states.
  • Backend/workflow: S3 Raw, EventBridge rule, SQS/DLQ, Step Functions execution history, Textract output, and AI Proxy logs.
  • Data: DynamoDB item, S3 Processed result.json, and sample API responses.
  • Ops/security: CloudWatch logs/alarms, X-Ray traces, SNS/SES email alerts, Secrets Manager, IAM summary, CloudTrail, and Budgets.

• Completed the 7-10 minute demo script:

  • Introduce the manual invoice/receipt processing pain point.
  • Present the high-level architecture and module ownership.
  • Demo login, invoice upload, workflow tracking, result view, and review case.
  • Present observability, alerting, cost guardrails, and cleanup.

• Reviewed the Definition of Done:

  • At least 3 sample invoices/receipts can be uploaded from the frontend.
  • Happy path, review path, and failure path all have evidence.
  • Textract extracts key fields; AI Proxy returns schema-valid JSON.
  • DynamoDB/S3 Processed store data using the agreed data contract and path convention.
  • CloudWatch/SNS/SES/Secrets Manager/Budgets/SAM cleanup all have evidence.

• Completed cleanup and cost control:

  • Prepared a manual cleanup checklist by service because many resources were created directly in the AWS Console.
  • Clarified that sam delete only applies to resources managed by a SAM/CloudFormation stack, not manually created resources.
  • Checked bucket objects, queues, tables, secrets, alarms, log groups, and notification resources in the AWS Console.
  • Reviewed AWS Budgets/Billing to reduce post-demo cost risk.