Features Overview

deployml is a Python library that deploys a complete MLOps infrastructure in GCP with a single command. It was built for academic use — the goal is to get the infrastructure out of the way so students can focus on the ML.

How it works

You define your stack in a YAML config file, run deployml deploy, and Terraform provisions everything in GCP. When you're done, deployml destroy tears it all down cleanly.

What gets deployed

Experiment Tracking, Artifact Storage, and Model Registry — MLflow

A single MLflow server running in Cloud Run, backed by: - Cloud SQL (Postgres) for experiment metadata - GCS bucket for model artifacts

Use it to track experiments, store models, and manage model versions.

Model Serving — FastAPI

A FastAPI container running in Cloud Run. On startup it pulls the latest registered model from MLflow and serves predictions at /predict.

Model Monitoring — Grafana

A Grafana container running in Cloud Run, connected to a metrics Postgres database. Use it to build dashboards for tracking model performance over time.

Feature and Monitoring Tables — BigQuery

Four BigQuery tables are created automatically in the mlops dataset:

Table	Purpose
offline_features	Precomputed input features for training and serving
predictions	Model predictions logged at serving time
ground_truth	Actual outcomes, matched back to predictions
drift_metrics	Summary statistics for monitoring model drift

What's not included (yet)

• AWS and Azure support (planned)
• Data versioning
• LLMs / generative AI
• Scalable model training