AI-Powered Transaction Monitoring for Miden 

Business Problem

Miden’s transaction monitoring process required significant manual effort, with employees scanning transaction logs, digesting analytics, and identifying anomalies. This manual approach led to: 

  • Operational inefficiencies due to high labor costs and time consumption. 
  • Delayed fraud detection, impacting financial security. 
  • Limited scalability, restricting Miden’s ability to handle growing transaction volumes effectively. 

About Miden

Miden is a modern issuer processor and banking stack that enables businesses to integrate financial features, including seamless payments, reconciliation, instant virtual cards, and account opening. They are on a mission to redefine financial transactions with enhanced security and efficiency. By leveraging AI, Miden aims to automate transaction monitoring and streamline operations. 

Explore Now

Solution

goML collaborated with Miden to develop an AI-powered Transaction Monitoring MVP, leveraging AWS’s advanced AI infrastructure to automate monitoring, enhance data usability, and improve system interaction through a conversational AI chatbot. This 10-week initiative significantly reduced manual efforts and improved operational efficiency. 

AI-Powered Transaction Monitoring for Miden
Click to View in Full Size

Architecture

  • User Interaction & API Layer 
    – Users interact with the system via an API Gateway, which handles queries and responses. 
    Lambda functions process user-based prompts and interface with downstream services. 
  • Data Processing & Storage 
    S3 stores Miden Data & Conversational Data
    – Data undergoes preprocessing before being used for further processing. 
    – Processed data is sent to Amazon Bedrock Anthropic Claude-v3.5 for question generation and Python code generation
    – Insights derived from the process are stored in Postgres for analytics. 
  • Model Inference & Search 
    – User-based prompts are processed through an Inference API
    – Data is retrieved from Postgres and embedded using Titan Embedding on EC2
    – Embedded data is indexed in OpenSearch for efficient querying. 
    – Amazon Bedrock Anthropic Claude-v3.5 is also used for inference. 
  • Deployment & Version Control
    – Code and models are deployed using GIT, ECR, and Docker.
    – Lambda functions and APIs are updated as part of the deployment pipeline.  
  • Security & Infrastructure Management 
    – IAM handles authentication and authorization. 
    – CloudFormation manages infrastructure as code (IaC) with Config Templates
    – CloudTrail logs activities for auditing and security compliance. 
  • Monitoring & Logging 
    – CloudWatch provides monitoring, CW Logs store logs, and CW Alarms trigger alerts. 
    – User Management is integrated with monitoring and logging for operational insights. 
     
Outcomes

0%

Reduction in interactive content creation by moving to AWS from GCP, From 2 days to less than 1 hour

0%

Improved Efficiency with
automated prompt engineering.

0%

Scalable Infrastructure and AWS infrastructure are leveraged to
handle increased demand

Technology Stack​