When developers think about the ideal AI-driven code review experience, they envision a tool that truly understands their code and delivers spot-on suggestions. Except what we see is that most tools struggle with high false-positive rates - accuracy rates are often below 20%. Even when they do flag issues correctly, the feedback can be trivial.
With our GA release one month ago, our customers are seeing continued code review accuracy with up to 60% accuracy in Baz-generated suggestions.
This is powered by our unique data model, which analyzes change requests post-merge and evaluates whether the merged code addressed our suggested changes. This continuous learning loop helps us refine our recommendations and ensure our feedback is not only accurate but also highly relevant.
The recent bundle of releases accelerates exactly that and takes advantage of the latest in RAG development via Voyage AI’s code embedding models.
February’s releases include:
- RAG Reviewer with deeper code insights and duplication detection
- Configurable Reviewer workflows tailored to your team’s needs
- Reviewer Agent collaboration to request more context in real time
Read more below on each new product feature. The Baz Reviewer is available and free to try here.
Unlocking New Code Insights with RAG
As your codebase and team scale, maintaining consistent coding conventions and adherence to established design patterns becomes increasingly challenging. With other review and Git tools, this has been tough to crack in the past because performing search across code bases were limited by lack of semantic understanding and accounting for framework specific conventions. Today with our RAG-powered reviewer, powered by Voyage AI, developers have a super power to quickly analyze codebases, extract code conventions, and automatically check them in every review. Interested in what this looks like behind the scenes? Here's a primer on how it works.
Baz Reviewer now analyzes your entire codebase to automatically generate your team's unique coding conventions. During every change request, it evaluates the new code against these conventions, ensuring alignment with your team's best practices. Baz Reviewer detects duplicate functionality and suggests reusing existing components, promoting code reuse, reducing technical debt, and improving maintainability. This is part of the overall Baz Reviewer workflows.
A Dynamic AI Companion Embedded in Your Change Request
Shortly after launching Baz Reviewer, we noticed users engaging by asking follow-up questions, providing additional context, and requesting examples to refine its suggestions. This type of interaction is exactly what we envisioned - a dynamic and helpful conversation around your code.
To support this, we quickly enabled follow-up interactions with Baz Reviewer’s findings. Now, you can simply reply to its comments in GitHub, and Baz Reviewer will intelligently determine whether it needs to step in with further assistance. This makes engaging with AI-generated suggestions much more collaborative and iterative, overall reducing friction and increasing your confidence in accepting AI-generated code.
Looking ahead, we’re working on allowing users to invite Baz Reviewer into any comment thread; not just those it initiates. This will empower teams to resolve issues faster and streamline code reviews even further.
Tailor Code Reviews to Your Needs
We believe that every CR should improve your codebase by adhering to all your review principles; from coding conventions and code hygiene to catching typos. However, we also recognize that some teams prefer a more flexible approach. That's why Baz Reviewer allows you to configure any of its workflows (with the exception of Breaking Changes and CI Errors, as these could introduce critical issues to your code).
Additionally, organizations can fine-tune reviews by excluding specific paths and file types, ensuring that the feedback you receive is both relevant and precise.
What’s Next for Baz AI Code Review
Developers deserve a code review process that leverages AI to get the most critical and most relevant changes first, catching major issues before the team has to bounce back and forth to review them. The Baz team is focused on building tools that understand not just your code but your software better. Our team is testing our tools and pushing the boundaries on code review every day.
Next on the roadmap includes the ability for developers to write their own customized workflows, and automated code principles based on team comments. We’re excited to share more soon and to hear your input on the code review process - what could be better? What tools are you using today?
.png)
.png)
