Components and Data Privacy¶

3LC is composed of the following main components:

  • The Python Package used to integrate with your existing ML code to capture metrics during training

  • The Object Service used to provide frontend components (Hub, Dashboard) access to ML training data and metrics captured during training, requires access to the same storage backend as the training

  • The Hub web application that serves as a no-code entry point to 3LC, enabling users to move seamlessly from data to model to insights without writing code or changing existing infrastructure, unifying dataset management, model training, performance analysis, and retraining into a single continuous, data-centric workflow

  • The Dashboard web application used to analyze and make revisions to ML training data and metrics in the browser, with ML data accessed through the Object Service

../_images/main-components.png

Data Privacy¶

The component architecture of 3LC has been intentionally designed to ensure that your ML data never leaves your infrastructure. Your ML data, including the metrics captured during training using the tlc Python Package, are stored in locations (local drive(s), network drive(s), and/or cloud storage) that you control. They are never uploaded to 3LC in any form.

The 3LC Object Service always runs on your own infrastructure, indexing your ML training data and metrics to make them available to the 3LC Hub and Dashboard.

The frontend web applications - the 3LC Hub and Dashboard - run in the browser on your own machine. They essentially need two things to allow you to analyze and edit your ML training data and metrics. First, they need the data itself, which they access through the Object Service running on your machine. Second, they need the static web resources (HTML and JavaScript) used to render their UIs, which they pull from the frontend servers (3LC Hub Frontend Server, 3LC Dashboard Server). Both frontend servers may be hosted by 3LC or in your own infrastructure depending on deployment, but in either case they are just the source of those static web resources and never receive any data.