Skip to Content (Press Enter)
Community › Forge › Assets list › AWS Claims Studio
Login to follow
AWS Claims Studio

AWS Claims Studio (ODC)

Stable version 0.1.3 (Compatible with ODC)
Uploaded on 25 August 2025 by Parth Sharma
AWS Claims Studio

AWS Claims Studio (ODC)

Details
Detailed Description

The same app (minor changes) used live on stage at OutSystems Developer Days Bengaluru 2025. It’s a simple example of how you can bring AWS AI services into an OutSystems app to automate insurance claims.


With AWS Claims Studio you can:

  • Upload a scanned claim form and vehicle photos.
  • Let Amazon Textract read and extract the form data automatically.
  • Use Amazon Comprehend to pick up sentiment from the form and notes.
  • Run Amazon Rekognition to identify and classify vehicle damage in photos.
  • Call AI Models (Claude 3.7 Sonnet on Bedrock for the demo) to generate a short claim assessment and next steps.
  • Get a clear Risk Score that ties everything together in one place.

Everything happens in a single, easy-to-follow screen.

Limitations
  • Built mainly as a demo app, not a production solution.
  • Handles one-page claim forms and image files only for now.
  • Uses heuristic mapping for image damage severity; not a trained model.
  • Risk scoring is illustrative and not based on actuarial rules.
  • Requires valid AWS credentials and access to Textract, Comprehend, Rekognition, and Bedrock; costs may apply.
  • Tested only with English forms and notes; other languages may produce inconsistent results.


Note:- The Textract functionality has been replaced from "AnalyzeDocument" to "DetectDocumentText" + static values, for forge submission purposes. On stage a custom .NET + Javascript + API workaround was used to populate the key value pairs. This is because the current AmazonTextract connector doesn't have a generic synchronous server call to AnalyzeDocument function. Implementing Async calls (StartDocumentAnalysis and GetDocumentAnalysis) is doable and even better for production, but were overkill (and we were also time bound) for the demo.