Veri is organized around a small set of surfaces: a web dashboard for managing work, an SDK and CLI for running from code, and a control plane that turns those requests into training, evaluation, and deployment jobs.
The goal is to keep model iteration close to a developer workflow. You should be able to connect data, start a run, inspect the result, and promote a checkpoint without managing schedulers, GPU providers, or one-off deployment scripts.
The current infrastructure is still intentionally narrow. The next step is making the loop feel more complete: clearer job history, better evaluation visibility, and a smoother path from an experiment to a hosted endpoint.