The road to smarter code reviewer recommendations

Jan 4, 2022 · 3 min read · Leave a comment
Taylor McCaslin GitLab profile

You may recall back in June 2021, we announced the acquisition of UnReview, a machine learning (ML) based solution for automatically identifying appropriate expert code reviewers and controlling review workloads and distribution of knowledge.

At the start of the new year we wanted to provide an update on our integration progress and our wider vision of leveraging machine learning to make GitLab's DevOps Platform smarter. You see, the acquisition of UnReview also was the initial staffing of our new ModelOps stage.

Our Newest DevOps Stage

This new stage, which we’ve named ModelOps, is focused on enabling and empowering data science workloads on GitLab. GitLab ModelOps aims to bring data science into GitLab both within existing features to make them smarter and more intelligent, but also empowering GitLab customers to build and integrate data science workloads within GitLab.

So what is ModelOps you may wonder? We view ModelOps as an all encompassing term to cover the entire end to end lifecycle of artificial intelligence models. We wanted to set our vision wide to fully cover everything needed to power data science workloads. DataOps is the processing of data workloads (think traditional ELT: extract, load, transform) and MLOps is the building, training, and deployment of machine learning models. If you’re confused don’t worry, it’s a lot to wrap your head around.

a look at the stages of MLOps

Today our DevOps Platform helps plan, build, test, secure, deploy, and monitor traditional software. Now we want to extend our DevOps Platform to include AI and ML workloads. If this is interesting to you, be sure to check out our recent Contribute talk where we dive deeper into plans for our ModelOps stage.

UnReview as our first feature

So what does this have to do with UnReview? Our acquisition of UnReview is going to be our first Applied ML group’s feature: suggested reviewers within GitLab’s existing reviewers experience. Today, a developer in a merge request has to manually choose a reviewer to look at their code. With UnReview we can leverage the contribution history for a project and recommend someone well-suited for code review of your specific changes.

Here’s an early mockup (and it may differ from our final UI) of how we’re thinking about this integration:

an early mockup of our UI

The UnReview algorithm looks at a variety of data points from your project’s contribution history to suggest an appropriate reviewer. We’re still in the early days of this integration but our initial internal testing shows great suggestions.

Customer beta coming soon!

This leads me to a final question, might you want to be one of our first customers to try this new code review experience? In early 2022, we’ll begin a private customer beta of this new functionality. If interested, fill out this form to express interest. Do note that we can’t accept everyone and we’ll focus initially on customer profiles that are well suited for the initial version of the suggestion algorithm. Our only ask is we’d like to find customers with active projects that have a healthy number of contributors. The model currently works best on larger repositories with lots of contributors where it may not immediately be clear who is an ideal code reviewer.

We can’t wait for customers to begin using this new reviewer suggestion experience and will be providing more updates in early 2022.

“Want better code review? Machine-learning-powered code review is coming to @gitlab and here's everything you need to know (and how you can help)” – Taylor McCaslin

Click to tweet

Open in Web IDE View source