AWS Lambda Fraud Detection POC

I am building a proof-of-concept that shows how AWS Lambda can detect fraudulent activity within a banking workload. The scope is limited to fraud detection and centers on two live data feeds: transaction history and user-behavior data.

The workflow I have in mind is event-driven: new transactions flow through Amazon Kinesis (or an SQS/SNS trigger if you prefer), land in Lambda, and are evaluated against a lightweight rules engine or ML inference endpoint. Any transaction flagged as suspicious should be written to DynamoDB (or another fast store) and surfaced through CloudWatch metrics and a simple API Gateway endpoint for review.

Key deliverables
• Well-commented Lambda function(s) in Python or Node.js that parse the incoming events, apply fraud-detection logic, and emit alerts
• Infrastructure-as-Code template (AWS SAM or CloudFormation) that deploys the full stack, including IAM roles
• README outlining setup steps, assumptions, and how to extend the rules/ML model
• Basic unit tests proving that normal and anomalous transactions are handled correctly
• Architecture diagram that maps data sources, Lambda, storage, and alerting flow

Acceptance criteria
1. A demo invocation using sample transaction and user-behavior payloads clearly shows fraudulent versus legitimate outcomes.
2. Deployment completes in one command with no manual console tweaks.
3. All resources may be torn down cleanly after testing.

Feel free to suggest alternate AWS services if they improve latency or cost, but the core of the POC must remain in AWS Lambda.