AI Reviewer for GitLab Merge Requests

• 

  Architecture

• 

  MR AI Reviewer in Action

• 

Inspiration

Manual code reviews are often slow, subjective, and create a bottleneck for fast-moving teams.
I wanted to automate the review process, reduce bugs, and empower developers to get instant feedback — powered by state-of-the-art AI.

What it does

mr-ai-reviewer is an AI-powered code review assistant for GitLab Merge Requests.
Every time a developer opens a Merge Request, the bot analyzes the code diff using Google Vertex AI, and posts line-by-line comments and a smart summary — directly in GitLab.

How I built it

• Integrated GitLab webhooks to trigger the workflow on every new MR.
• Parsed the code diff and built prompts for Google Vertex AI (Gemini).
• Used Vertex AI to analyze code quality, security, and best practices.
• Posted AI-generated feedback and summaries via GitLab API.
• Deployed the solution on Google Cloud Run for scalability and reliability.
• Created demo and documentation for easy testing.

Challenges I ran into

• Mapping AI feedback to the correct lines/files in multi-language diffs..
• Making the bot non-intrusive (no spam, only actionable feedback).

Accomplishments that I'm proud of

• Fully automated, end-to-end AI code review for any GitLab project.
• Seamless inline comments and summary — even for large diffs and multiple languages.
• Production-ready Docker images (ARM64/AMD64) and public Cloud Run demo.
• Friendly README and public demo repo for anyone to try.

What's next

• Per-user and per-team analytics dashboards.
• Enhanced customization for team/company code style and review policies.
• Vector database and memory integration:
  Store and track previously reviewed Merge Requests, so the AI can reference past feedback, identify recurring issues, and provide more context-aware, personalized reviews.

Built With

• docker
• fastapi
• gitlab-api
• google-cloud-run
• google-vertex-ai-(gemini)
• pytest
• python