Kubernetes to serve a fraud detection model
summary
Claims a deployed real-time fraud API cuts fraud ~50% in six months and raises valid conversion ~4% while serving subsecond decisions under load. Architecture uses Python Flask containers on EKS, Docker images in ECR, Terraform for infra, Helm for zero-downtime deploys, Horizontal Pod Autoscaler on CPU, preStop hooks and automated rollbacks driven by end‑to‑end CI tests. Feature pipeline uses Kinesis/SQS, EMR jobs, Redis Elasticache online store and RDS offline storage and it scales to handle promotion spikes without latency degradation.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:18How credit card fraud works
- 2:19Chargebacks
- 5:44Machine Learning
- 7:08ML Features
- 8:54ML Training
- 9:32Order Evaluation API
- 11:44Building the API
- 13:53Autoscaling
- 16:40Automated test suites
related talks
51mReal-time Fraud Detection: Challenges and Solutions
25mReal-Time Fraud Detection at Scale
07mScaling Predictive Manufacturing Analytics: Kubernetes-Powered CRM
13mBuilding Financial Fortresses: Real-Time Fraud Detection
15mScaling AI-Driven Financial Services: Ethical ML at Scale
10mOptimizing Machine Learning (ML Ops) in Financial Services with Kubernetes and Cloud Technologies.
14mPerformance Optimization in High-Volume Transaction Systems
13mBuild Real-Time Analytics Platforms for Financial Crime Detection