Overview
A maintainable MLOps workflow for demand forecasting that separates experiment tracking, model promotion, batch inference, and monitoring.
Problem
Forecasts were created manually and were difficult to reproduce, compare, or deploy consistently.
Solution
Created a modular forecasting system with repeatable training, tracked metrics, scheduled scoring, and API-ready predictions.
Architecture
How the system is structured
01
Training jobs write metrics and artifacts to MLflow
02
Batch scoring publishes forecasts to analytics tables
03
FastAPI service exposes selected forecast endpoints
04
Docker Compose setup keeps local development reproducible
Tech Stack
PythonMLflowFastAPIDockerPostgreSQL
Key Features
Experiment trackingModel promotion workflowBatch and API inference pathsMonitoring-ready output tables
Screenshots
Visual walkthrough
MLflow comparison
Forecast API
Batch scoring logs