IaC (Infrastructure-as-Code) code to provision resources for a minimalistic ML platform. The resources provisioned are:

• A Postgres database
  • Stores information such as: experiment IDs & names, run metadata (start time, end time, status), logged metrics, etc.
• An Azure Blob Container
  • Stores the actual trained model files & artifacts
• An Azure Kubernetes Cluster
  • To deploy models for inference

Install Pulumi first from here: https://www.pulumi.com/docs/iac/download-install/

Then git clone this repo, and run pulumi up --yes --skip-preview to set up all the defined resources on Azure.

• Feel free to change up the code in index.ts to suit your needs. For example, you might want more than 1 agentpool and more than 1 node per agentpool. I am using one for each to save Azure costs.

See previous experiments I ran: http://minimlplatform.duckdns.org/mlflow

• If the link does not work, it is probably because I paused my Kubernetes cluster so I don't go broke

Train & Deploy your own model from here: https://github.com/EddieJ03/model-template

• Example of models I have created and deployed:
  • https://github.com/EddieJ03/review-predictions-model
  • https://github.com/EddieJ03/translation-model