ML Platform Stack
From experiment to production model in a single coherent platform
Building an ML platform from scratch? This stack covers the full lifecycle: data versioning, experiment tracking, feature management, distributed training, model serving, and monitoring. Used by ML engineering teams moving beyond ad-hoc notebooks.
DVC
Git for data and models. Tracks datasets and model artifacts alongside your code without bloating the Git repo.
MLflow
Logs parameters, metrics, and artifacts for every training run. Model registry keeps production models auditable.
Feast
Decouples feature engineering from model training. Serves the same features online (low-latency) and offline (batch).
Ray
Scales Python ML workloads across a cluster. Works with PyTorch, TensorFlow, and XGBoost without changing your code.
BentoML
Packages models into production-ready APIs with batching, adaptive scaling, and multi-model serving built in.
Evidently AI
Detects data drift, target drift, and prediction quality degradation in production. Open-source with a clean dashboard.