MLOps

Getting models out of notebooks and keeping them healthy in production.

1. Step 01

   Engineering baseline

   Python packaging, testing, Git workflow and Docker for reproducible environments.

   PythonDocker

2. Step 02

   Data pipelines

   Ingestion, validation, versioned datasets and scheduled transformation jobs.

   Airflowdbt

3. Step 03

   Experiment tracking

   Runs, parameters, artefacts and a model registry with promotion stages.

   MLflowW&B

4. Step 04

   Serving

   REST and batch inference, GPU vs CPU trade-offs, autoscaling and warm starts.

   FastAPITriton

5. Step 05

   Monitoring

   Data drift, concept drift, latency and quality dashboards with alerting.

   EvidentlyPrometheus

6. Step 06

   Automation

   CI/CD for models, automated retraining, rollback and shadow deployments.

   CI/CDKubernetes