How fuzzylabs/matcha works
Project type and purpose
• Matcha is an open-source Python CLI tool for one-step MLOps provisioning on Azure.
• Its purpose is to let data scientists and ML engineers provision an end-to-end ML infrastructure (experiment tracking, model training, deployment) using Terraform, ZenML, and Azure services.
File structure
• Top-level:
• src/matcha_ml: the Python package
• tests: unit tests
• docs: MkDocs-based documentation
• .github, CI/CD workflows, pre-commit, linters
• src/matcha_ml/cli: command-line interface and UI logic
• src/matcha_ml/core: core orchestration and validation
• src/matcha_ml/config: matcha_config parsing and defaults
• src/matcha_ml/services: abstractions around Terraform, Azure, analytics
• src/matcha_ml/runners: execution layers for local/remote/azure runs
• src/matcha_ml/templates: Terraform template generators
• src/matcha_ml/infrastructure: pre-packaged Terraform modules for different “stacks” – default (MLflow, AKS, ACR, storage, Seldon, ZenML), llm (Chroma, LLM stack), remote_state_storage
• src/matcha_ml/state & storage: state handling (remote/local) and Azure storage wrappers
README for additional insights
• Confirms MLOps stack includes experiment tracking, training, deployment
• References ZenML for stack recipes and various Azure permission requirements