How Sisense built its best review practices into every merge request with CodeRabbit
CodeRabbit helps Sisense apply its engineering conventions across their repositories, catch issues earlier, and give human reviewers more time to focus on logic and architecture.
CASE STUDYEngineering standards applied with more than 3,100 CodeRabbit learnings
As Path instructions accelerated, Functional Correctness hit 73%
More human attention for logic and architecture
12.3 weeks of reviewer time saved
Sisense provides an enterprise-grade business intelligence and data analytics platform that helps organizations collect, analyze, and visualize complex data. Its engineering organization includes approximately 80–85 developers working across frontend, backend, and DevOps teams.
Before CodeRabbit, Sisense conducted code reviews manually in GitLab. Thorough reviews demanded time and concentration—two resources that became harder to protect when teams were under pressure to deliver. Reviewers could miss defects, allowing costly bugs and security issues to reach production. They also spent valuable time catching smaller concerns such as incorrect types, missing guards, naming inconsistencies, and uncovered edge cases.
“We wanted to lift this burden from the developer,” said Bogdan Karlenko, Engineering Manager at Sisense.
Challenge: Keep human review focused on meaningful decisions
Sisense initially adopted CodeRabbit before AI agents were generating a significant share of its code. GitHub Copilot was helping developers write code faster, but most changes were still human-directed.
Even then, CodeRabbit gave the team a way to automate repetitive review work. It could identify type problems, incomplete test coverage, unsafe assumptions, and other code-level concerns before a senior developer entered the conversation.
“When a human looks through naming, then starts suggesting different names, then sees some edge cases that aren’t covered, they get tired,” Bogdan said. “They either approve it as is, postpone it until the next day, or the whole review process becomes slower.”
The challenge grew as Sisense adopted Cursor, Claude Code, Codex, and other AI development tools. Engineers could produce more code faster, but human review capacity did not increase at the same rate.
“Human review is still a bottleneck, and we want the human in the middle,” Bogdan said. “We’re not ready to push everything into master and release it before people look through it.”
Solution: Put CodeRabbit first in the merge-request workflow
Sisense evaluated other AI review tools, but found that some lacked the integration or capabilities it needed. CodeRabbit offered the strongest balance of features and value while fitting directly into the company’s existing software development lifecycle.
Its integration with Sisense’s self-hosted GitLab instance was particularly important. CodeRabbit participates in the merge-request workflow like another team member, allowing Sisense to preserve existing requirements, such as resolving comments before merging.
CodeRabbit’s automatically generated summaries and sequence diagrams also help reviewers understand complex merge requests more quickly. After testing the platform, Sisense surveyed its developers and asked whether the team should keep it. A majority voted yes.
“Some edge cases were pointed out, and that was amazing, especially for less mature developers,” Bogdan said. “It can be automated before the senior developer gets to the review, so it saves us multiple loops.”
Connecting CodeRabbit to self-hosted GitLab
Connecting CodeRabbit to Sisense’s self-hosted GitLab environment required the company’s IT team to establish a secure channel through its VPN. Once connected, Sisense began adding repositories and saw CodeRabbit generate useful feedback immediately.
The CodeRabbit team also responded quickly when Sisense encountered issues. When Sisense requested better dashboard visibility into adoption and software development metrics, an improved dashboard containing its requested capabilities arrived within days.
CodeRabbit = Sisense review practices in every merge request
Before CodeRabbit
- Human reviewers spent time on types, naming, missing guards, test coverage, and edge cases.
- Review quality depended on scarce time and concentration.
- Internal conventions were difficult to apply consistently across repositories.
With CodeRabbit
- Every merge request receives an immediate, consistent first pass.
- Guidance set per repository, folder, and package carries Sisense conventions into each review.
- Human reviewers focus more attention on logic, architecture, structure, and feature readiness.
Turning 2,000 human reviews into an AI review system
As AI coding tools became more widely adopted, Sisense explored whether it could reproduce the behavior of its strongest human reviewers internally by building its own code-review tool.
The team analyzed approximately 2,000 past merge requests and organized the feedback into 13 categories and 62 subcategories. It then created several AI reviewer personas that mimicked different review behaviors and attempted to have them evaluate changes and reach consensus.
“It became hectic,” Bogdan said. “Those personas were conflicting and doing multiple runs. It was entertaining and occasionally insightful, but it proved to be more fun than productive.”
The experiment still produced a valuable asset: a structured record of the conventions and concerns that guided Sisense’s best reviewers. Rather than continue maintaining a custom multi-agent system, Sisense incorporated that knowledge into CodeRabbit.
Why CodeRabbit stood out
Repository-specific reviews grounded in Sisense conventions
Sisense created CodeRabbit configuration files for individual repositories, enhanced the review instructions, and added guidance for folders and packages within its frontend monorepo. Those instructions tell CodeRabbit what each package contains, why it exists, how it interacts with other packages, and which internal conventions apply to it.
“Because CodeRabbit allows us to configure it per repository, that worked perfectly,” Bogdan said. “We added per-folder and per-package instructions. Because we use a monorepo, that was very useful.”
CodeRabbit can now ask whether a developer should extend an existing factory instead of implementing something from scratch. It can question whether a component should be public, recommend moving shared behavior into a utility, or align names with internal conventions.
“CodeRabbit takes care of best practices in the code and best practices in the repository—the very specific, internal things,” Bogdan said. “It now catches issues that generic AI reviewers cannot catch or may miss.”
Broad coverage across the technology stack
Across Sisense’s stack, CodeRabbit identifies issues ranging from security and application logic to configuration and tooling. Its findings have included duplicate functionality, unnecessary API exposure, missing error handling, overlooked edge cases, and configuration concerns involving npm, Yarn, and Helm. It can provide feedback across Node.js, Java, Python, Groovy, and Bash code.
Within the frontend monorepo, CodeRabbit is especially useful for TypeScript. It can identify incomplete unions, unexpected variable types or values, unsafe property access, and unit tests that fail to cover relevant cases.
“It is good at finding edge cases. It is also good at reviewing unit tests. It can point out that something is not covered or that a test is not covering everything.”
Bogdan Karlenko, Engineering Manager at Sisense
Results: Faster reviews with fewer escaped bugs
Most importantly, CodeRabbit’s checks happen before a human reviewer spends time on the review. Senior developers can concentrate on business logic, architecture, structure, and feature readiness.
“Our reviewers can look through the logic and the structure and communicate interpersonally without needing to catch all those other issues,” Bogdan said.
Sisense still requires a human developer to review and approve changes. CodeRabbit strengthens that process by providing an immediate, consistent first pass. Developers spend less time on repetitive review work and gain confidence knowing CodeRabbit has already examined the change.
“We can review faster, which means we can ship faster and fix bugs faster,” Bogdan said. “Our quality is stable, and we have fewer escaped bugs.”
A quiet reviewer embedded in the workflow
Sisense primarily uses CodeRabbit through its CI/CD process, with additional IDE integration available for developers who want feedback before pushing code to GitLab. Bogdan values that CodeRabbit operates in the background instead of introducing another competing interface into an increasingly crowded AI development workflow.
“CodeRabbit for us is a silent reviewer, and that’s great,” he said. “It’s not in the way, but it is very valuable. It is a team member for us.”
The team configured CodeRabbit to suppress less-important output and keep meaningful findings prominent. Developers can manage those settings alongside the code without relying on an administrator.
“It will remind you when it is needed,” Bogdan said, “but it will not bother you when you are in the flow.”
As Sisense increases its use of AI coding agents, CodeRabbit provides an independent review layer that helps the company turn greater development speed into reliable software.
“Coupling it with something that is writing code can help us deliver more, faster, and with higher quality,” Bogdan said. “We really got used to CodeRabbit. It is very comfortable, and it produces value.”
New York, USA and Israel
https://www.sisense.com/Languages
TypeScript, Java, Python, Groovy, Bash
Challenge
Scale AI-assisted development while maintaining consistent review standards and code quality across their repositories.
