Teams building and operating ML models often live with a recurring challenge: the artefacts that make a model useful — weights, training data, experiment code, configs, metadata — are scattered across notebooks, storage buckets, trackers, and scripts. This friction shows up as last-minute fixes, delivery delays, and handoffs that feel like guesswork.
ModelPack, a community-driven specification from the Cloud Native Computing Foundation, defines a simple, open way to package and share ML projects so they behave like any other versioned asset in your toolchain.
In this session you’ll hear how KitOps, the most widely adopted implementation of the ModelPack standard, supports data scientists and engineers to: create reproducible packages, keep everything versioned in registries they already use, and automate pipelines without reinventing infrastructure. We’ll explore real-world flows where packaging replaces guesswork, tools interoperate more predictably, and collaboration between science, engineering, and operations feels grounded in common practices rather than bespoke scripts