MLOps Project

Overview

Course project applying end-to-end MLOps practices (reproducibility, CI, Docker, deployment, monitoring) to a machine-learning pipeline, with the tabular CNN baseline being replaced by an MLP.

Built on the DTU MLOps cookiecutter template with a structured repo (configs, data, dockerfiles, tests, reports, docs). Focused on reproducibility, automated testing, containerized training/inference, and a FastAPI service while migrating the tabular model from CNN to MLP.

What I learned

• • Structured an MLOps repository with configs, data splits, tests, reports, and documentation
• • Set up reproducible environments and dependency management for ML workflows
• • Versioned data and experiments for traceability and rollback
• • Built CI pipelines for testing, linting, and coverage
• • Containerized training and API services with Docker
• • Designed an inference API with FastAPI for deployment-ready integration
• • Improved model fit for tabular data by moving from CNN to MLP