ModelBox is a new AI experimentation and model operations service. AI Metadata services are integrated mainly into larger platforms and impose a specific workflow on the users. Most of the tools in this space are also SAAS services, such as HuggingFace, Neptune AI, etc.
The number of open source services that provide the primitives of logging metadata from experiments, model store/registry are few, and to mention services that are hackable and extendable to suit the needs of an organization with a specific workflow. We try to change that with ModelBox, by building the service with an interface first approach and having implementations that use specific storage technologies, etc. The API of ModelBox uses gRPC, with SDKs built around the API in the most popular languages used by engineers making AI services today.
In the coming weeks or months, we will write more about how ModelBox can be used to solve common use cases while developing models and help with serving and evaluating trained models.
Please get in touch with us on the GitHub Discussions or Discord if you want to be involved with the project!