ai-diagrams · Diagram examples · Sample fintechTransfer fraud prevention · V1
■  AWS architecture poster

Real-time fraud prevention for transfers

The risk decision runs within the transfer request, with a 300 ms budget at p99. The decision Lambda reads features from DynamoDB, requests a score from SageMaker and publishes the decision to Kinesis; two consumers update the features and alert the analyst.
AWS Cloud
Synchronous decision
Asynchronous follow-up
Customer app
outside AWS
API Gateway
POST /transferencias
Decision Lambda
applies the score band
DynamoDB features
aggregates per account
SageMaker endpoint
risk model, score 0 to 1
Kinesis Data Streams
decision recorded v1
Features Lambda
recalculates aggregates
Alert Lambda
filters review and block
SNS
cases for review
Fraud analyst
outside AWS
1
2
3
4
5
6
7
8
9
10
Steps
1
The customer app calls POST /transferencias with the session token.
2
API Gateway invokes the Decision Lambda synchronously; the request budget is 300 ms at p99.
3
The Lambda reads from DynamoDB features the account and recipient aggregates: average amount over 30 days, new recipients in 24 hours, key age.
4
The SageMaker endpoint receives the feature vector and returns a score from 0 to 1.
5
The Lambda returns approve, review or block to the app and publishes decision recorded v1 to Kinesis, using the account as the partition key.
6
The Features Lambda consumes the stream in account order.
7
The recalculated aggregates return to DynamoDB features and apply to the next transfer.
8
The Alert Lambda reads the same stream and filters the decisions review and block.
9
The case is published to the SNS topic for review cases.
10
The fraud analyst receives the case and decides in the internal console; a transfer under review is held for up to 30 minutes.
Diagrams  ·  AWS architecture example  ·  2026-11