Back to Projects
Data EngineeringFeatured

Credit Scoring Batch API

Production FastAPI microservice for batch ingestion of 18+ financial and behavioral data domains, powering credit scoring for 5 Ethiopian partner banks via Feast Feature Store and ML pipelines

5–10 pod autoscaling under load
Performance
99.9% pipeline reliability
Improvement
Serving 5 Ethiopian partner banks
Impact

Tech Stack

FastAPIMongoDBFeast Feature StoreKubernetes (EKS)DockerAWSPrometheusKubeflow PipelinesPythonPyMongoPandasGitHub Actions

Problem Statement

Five Ethiopian partner banks needed a reliable, scalable API to ingest diverse financial and behavioral data — transactions, SMS, KYC, psychometrics, employment, demographics, invoices — for ML-driven credit scoring and fraud detection.

Technical Approach

Built a FastAPI microservice with 22+ REST endpoints across 18+ data domains, MongoDB backend with 25 collections, and Feast Feature Store integration for ML feature pipelines. Deployed on AWS EKS with Horizontal Pod Autoscaler (5–10 replicas), CI/CD via GitHub Actions, and Prometheus metrics for observability.

Key Results

  • 22+ REST endpoints across 18+ data domains (SMS, KYC, transactions, psychometrics, employment, etc.)
  • Feast Feature Store integration enabling credit scoring and fraud detection ML models
  • AWS EKS deployment with HPA auto-scaling (5–10 pods) and zero-downtime rolling updates
  • CI/CD pipeline via GitHub Actions with automated ECR builds and EKS rollouts
  • Prometheus metrics for real-time monitoring of MongoDB ops, auth, and domain request rates