How aakash1552005/telco-churn-mlops works
This is a Telco customer-churn MLOps platform: it ingests and validates data, engineers features, trains and evaluates models, applies promotion governance, then serves predictions through a secured API. Its operational loop adds drift detection, monitoring, and automated retraining. The graph includes the documented MLflow, Prometheus/Grafana, Kubernetes, and Jenkins capabilities; their specific runtime wiring is not fully shown in the excerpts, so those links are documented rather than inferred from co-location.
Main entry points: routes.py API routes · app.py FastAPI application · drift.py Drift detection · metrics.py Metrics instrumentation
Components 24
Data preparation
- Data ingestion
src/data/ingestion.py - Schema validation
src/data/validation.py - Feature engineering
src/data/features.py
Model lifecycle
- Model training
src/training/train.py - Model evaluation
src/training/evaluate.py - Promotion policy
src/training/promotion.py - MLflow registry
Prediction service
- FastAPI application
src/api/app.py - API routes
src/api/routes.py - Request schemas
src/api/schemas.py - API security
src/api/security.py - Prediction service
src/inference/service.py
Monitoring and retraining
- Metrics instrumentation
src/api/metrics.py - Drift detection
src/monitoring/drift.py - Drift state
src/monitoring/state.py - Retraining trigger
src/monitoring/jenkins_trigger.py
Runtime platform
- Configuration
src/core/config.py - Structured logging
src/core/logging.py - Prometheus
- Grafana
- Kubernetes runtime
infra/k8s/deployment.yaml - Jenkins
Other components
- Dataset source
- API client
Connections 22
- Dataset source → Data ingestion: supplies data
src/data/ingestion.py - Data ingestion → Schema validation: validates dataset
- Schema validation → Feature engineering: checks schema
- Feature engineering → Model training: prepares features
- Model training → Model evaluation: evaluates model
src/training/evaluate.py - Model evaluation → Promotion policy: informs promotion
- Promotion policy → MLflow registry: registers model
src/training/promotion.py - MLflow registry → Prediction service: loads production model
src/inference/service.py - API client → FastAPI application: sends requests
- FastAPI application → API routes: includes routes
src/api/app.py - API routes → Prediction service: requests prediction
src/api/routes.py - API routes → Request schemas: validates payloads
src/api/routes.py - API routes → API security: applies security
src/api/routes.py - API routes → Metrics instrumentation: records telemetry
src/api/routes.py - Drift detection → Drift state: tracks windows
src/monitoring/drift.py - Drift detection → Retraining trigger: triggers retraining
src/monitoring/drift.py - Drift detection → Metrics instrumentation: updates drift metrics
src/monitoring/drift.py - FastAPI application → Metrics instrumentation: sets model info
src/api/app.py - Prometheus → FastAPI application: scrapes metrics
README.md - Grafana → Prometheus: visualizes telemetry
README.md - Kubernetes runtime → FastAPI application: orchestrates serving
- Jenkins → Retraining trigger: runs trigger