ml-prod 2024
projects
A hands-on tour of taking a machine-learning model all the way to production — from a trained model to a containerized, orchestrated, monitored service — built around the IBM AI Enterprise Workflow curriculum. The recurring theme is the full lifecycle: data → train → serve → containerize → orchestrate → monitor → retrain.
What’s inside
- Model microservice — a Flask API exposing train and predict endpoints, with versioned model artifacts, structured logging, and a monitoring module that watches for data drift. Packaged in Docker, with unit tests covering the API, the model, and the logger.
- Kubernetes — deploying and scaling the service on a cluster.
- AI Workflow capstone — time-series revenue forecasting from raw invoice data: a data-ingestion pipeline that reconciles messy JSON, exploratory analysis, and a comparison of forecasting models.
- Deployment solution — tying the pieces together into a deployable workflow.
Why it matters
Most of the work in an ML system isn’t the model — it’s everything around it: serving predictions reliably, shipping reproducible containers, scaling under load, logging and monitoring for drift, and closing the loop by retraining when performance degrades. This project walks through each of those stages on a concrete example.