How karthikk022/mlops-k8s-platform works
MLOps K8s Platform automates model development and operations on Amazon EKS. The principal workflow generates or supplies data, validates and preprocesses it, trains and evaluates a model, tracks and registers it in MLflow, then serves it through KServe with canary rollout. Separate branches cover Feast feature serving, drift monitoring and observability, infrastructure and security, and disaster recovery. Sampled code confirms pipeline stage calls, model loading, feature materialization, and drift metric export; other links reflect README documentation, not inspected implementation.
Main entry points: pipeline.py Pipeline Orchestrator · InferenceService.yaml KServe Deployment · handler.py Inference Handler · terraform EKS Infrastructure
Components 22
ML Lifecycle
- Sample Data
data/generate_sample_data.py - Pipeline Orchestrator
ml-pipeline/pipeline.py - Data Validation
ml-pipeline/pipeline.py - Preprocessing
ml-pipeline/preprocessing/preprocess.py - Model Training
ml-pipeline/training/train.py - Evaluation Gate
ml-pipeline/evaluation/evaluate.py
Model Platform
- MLflow Registry
infrastructure/helm/mlflow - Feature Definitions
feature-store/feature_definitions.py - Feature Materialization
feature-store/materialize.py - Inference Handler
serving/model-server/handler.py - KServe Deployment
serving/InferenceService.yaml - Canary Rollout
serving/canary-rollout.yaml
Observability
- Drift Monitor
monitoring/evidently/drift_monitor.py - Prometheus Alerts
monitoring/alerts - Grafana Dashboards
monitoring/dashboards
Infrastructure
- EKS Infrastructure
infrastructure/terraform - Admission Policies
security/gatekeeper
Security and Recovery
- Cluster Backups
backup-dr/velero - RDS Snapshots
backup-dr/terraform/rds-snapshots.tf - S3 Replication
backup-dr/terraform/s3-replication.tf
Other components
- ML Engineer
- AWS Services
Connections 21
- ML Engineer → Sample Data: generates data
README.md - ML Engineer → Pipeline Orchestrator: runs pipeline
README.md - Pipeline Orchestrator → Data Validation: calls validation
ml-pipeline/pipeline.py - Pipeline Orchestrator → Preprocessing: calls preprocessing
ml-pipeline/pipeline.py - Pipeline Orchestrator → Model Training: calls training
ml-pipeline/pipeline.py - Pipeline Orchestrator → Evaluation Gate: calls evaluation
ml-pipeline/pipeline.py - Model Training → MLflow Registry: logs and registers
ml-pipeline/training/train.py - Evaluation Gate → MLflow Registry: loads model
ml-pipeline/evaluation/evaluate.py - MLflow Registry → Inference Handler: supplies model
serving/model-server/handler.py - Inference Handler → KServe Deployment: serves inference
README.md - KServe Deployment → Canary Rollout: routes rollout
README.md - Feature Definitions → Feature Materialization: defines features
README.md - Feature Materialization → KServe Deployment: serves online features
README.md - Inference Handler → Drift Monitor: produces predictions
- Drift Monitor → Prometheus Alerts: exports metrics
monitoring/evidently/drift_monitor.py - Prometheus Alerts → Grafana Dashboards: visualizes metrics
README.md - EKS Infrastructure → KServe Deployment: hosts serving
README.md - Admission Policies → EKS Infrastructure: enforces policies
README.md - EKS Infrastructure → Cluster Backups: backs up cluster
README.md - AWS Services → RDS Snapshots: stores snapshots
README.md - AWS Services → S3 Replication: replicates objects
README.md