We wanted to implement DevOps best practices for AI projects to make sure that software engineers and data scientists could work together seamlessly building solutions.

What it does

A tutorial on how to build CI using GitHub Actions workflows or/and Azure DevOps pipelines for models built with We built testing framework to run batch tests for and developed CLI (wittycli) for this tutorial. This tutorial helps software engineers and AI developers to follow engineering best practices like continuous integration and deployment and test driven development, pull requests reviews and apply these practices to models.

How we built it

We used GitHub Actions and Azure DevOps and our own CLI to build workflows for testing and developing models on and we've written a tutorial about how to do it.

Challenges we ran into

There is no official CLI for, so we've built it ourselves . There is no official "unit testing"/batch testing system for to validate trained model version, so we developed the approach of how as a developer you can test model programmatically by comparing outputs with expected intents and entities and confidence levels.

Accomplishments that we're proud of

  • First CI pipeline for models.
  • Our own CLI that's easily extendable and a pleasure to use.
  • model testing framework.

What we learned

A bit about how APIs work and what to avoid when using it.

What's next for CI and CLI

  • Add support for more API calls from the REST API in wittycli
  • Add more examples

Built With

Share this project: