Mentor MR : Learn from GitLab's best , before you merge

MentorMR is an AI tool that learns from GitLab's top contributors to provide personalized code review feedback. It helps new contributors feel confident and improves feedback quality for developers.

Comment

Inspiration

🎯 MentorMR: My Solo Hackathon Journey

The inspiration for MentorMR came from a simple observation: the best learning happens during code reviews, but most developers never get feedback from world-class experts. I wondered: "What if I could use AI to give every developer code reviews styled after GitLab's co-founder Dmitriy Zaporozhets or CEO Sid Sijbrandij?" When I saw the GitLab + Google Cloud Hackathon 2025, I realized this was my chance to democratize expert-level code reviews using Google's Gemini AI.

What it does

MentorMR represents a new category: Expertise Amplification Systems. Instead of replacing human experts, it amplifies their knowledge and makes it accessible to every developer. I envision a future where junior developers get feedback styled after senior engineers, open source mentorship scales infinitely, and AI serves as a bridge between experts and learners. Innovation: First AI that learns from specific GitLab contributors rather than providing generic feedback. Technical Excellence: Production-ready deployment using Gemini 1.5 Flash on Google Cloud with real GitLab API integration. Transparency: Complete openness about data sources and AI processing—users can verify everything. Real Impact: Transforms every code review into a learning opportunity mentored by the people who built GitLab itself.

How we built it

#Architecture Decision

I chose a serverless approach with Google Cloud Functions + Gemini AI + real GitLab data:

GitLab Webhook → Cloud Function → Gemini AI → Personalized Review

Core Components

  1. GitLab Analyzer - Fetches real contributor data from GitLab CE repository (278,964+ commits analyzed)
  2. AI Reviewer - Uses Gemini 1.5 Flash to generate reviews in contributor styles
  3. Webhook Handler - Processes live GitLab events and orchestrates responses
  4. Demo Interface - Interactive web demo showcasing real-time AI reviews

The AI Breakthrough

The key innovation was creating prompts that make Gemini understand contributor personalities:

prompt = f"""You are {contributor_name} reviewing code.
Style: {review_patterns}
Focus: {expertise_areas} 
Sample: "{actual_review_comment}"
Review this: {code_diff}"""

Challenges we ran into

1. Making AI Sound Human

Challenge: Generic AI feedback is forgettable. Solution: Analyzed actual review comments from GitLab's top contributors and fed their communication patterns into Gemini's prompts.

2. Proving Real Data Usage

Challenge: Many hackathon projects use fake data. Solution: Built live API validation endpoints that prove MentorMR uses authentic GitLab contributor data, not mock information.

3. Production-Ready in Days

Challenge: Building something teams could actually use, not just a demo. Solution: Focused on proper error handling, scalable Google Cloud deployment, and simple GitLab integration via webhook URLs.

4. API Rate Limiting

Challenge: GitLab API has strict rate limits for data collection. Solution: Implemented intelligent caching and focused analysis on GitLab's top 20 contributors for maximum impact.

5. Real-time Performance

Challenge: Users expect instant feedback, not slow AI processing. Solution: Optimized Gemini prompts and used Cloud Functions for fast cold starts and warm request handling.

Accomplishments that we're proud of

  • Personalized AI beating generic AI: Reviews styled after real experts are far more impactful than generic feedback
  • Data authenticity matters: Using real GitLab contributor data instead of mock data creates genuine value
  • Transparency builds trust: Making the AI process visible and verifiable increases user confidence

What we learned

Technical Mastery

  • Gemini AI Integration: Discovered how to make Gemini 1.5 Flash understand and mimic specific contributor personalities
  • GitLab API Deep Dive: Learned to extract meaningful patterns from real contributor data and review styles
  • Production AI Deployment: Built my first production-ready AI system on Google Cloud Functions
  • Real-time Webhooks: Mastered GitLab webhook integration for live code review automation

What's next for Mentor MR :

Expand across all GitLab projects to analyze contributors from popular repositories like GitLab Runner, GitLab Pages, and other ecosystem projects, creating specialized AI mentors for different domains. Build enterprise GitLab integration by allowing companies to create custom MentorMR profiles from their internal GitLab contributors, preserving and scaling their senior developers' expertise. Launch as a native GitLab feature through official partnership, making AI-powered mentorship available to every developer in the GitLab community.

Built With

  • ai/ml-stack-gemini-1.5-flash
  • cloudfunction
  • cloudshell
  • contributorapi
  • flask
  • gcp
  • generativemodel
  • gitlabcerepository
  • gitlabrestapiv4
  • googlecloudsdk
  • html5
  • javascript
  • javascript-(frontend)
  • livegitlabapiendpoints
  • projectid278964
  • python-(backend)
  • python-dotenv
  • storage
  • vertex-ai
  • webhooks
Share this project:

Updates