Somm.dev

Six AI sommeliers taste your code. Powered by Gemini 3, they evaluate architecture, quality, and security using 72 techniques — delivering a verdict in under a minute.

Comment

Inspiration

The Gemini 3 Hackathon had over 30,000 participants and only 17 days to judge. That's impossible for humans to review fairly.

We asked: What if AI could evaluate code the way a wine sommelier tastes wine?

What it does

Somm.dev is an AI-powered code evaluation platform where 6 autonomous agents review your code:

  • 🍷 Jean-Pierre — Grand Sommelier, Final verdict
  • 🏛️ Marcel — Architecture & design patterns
  • ✨ Isabella — Code quality & maintainability
  • 🔒 Heinrich — Security analysis
  • ⚡ Laurent — Performance optimization
  • 💡 Sofia — Innovation & best practices

Each agent applies 72 evaluation techniques, powered by Gemini 3's 1M token context.

How we built it

  • Frontend: Next.js, React, TailwindCSS
  • Backend: Python, FastAPI
  • AI Orchestration: LangGraph
  • Core AI: Gemini 3 Pro (1M context + RAG)
  • Visualization: 3D agent interaction graph

Challenges we ran into

  • Coordinating 6 parallel agents without conflicts
  • Keeping evaluation consistent across different codebases
  • Visualizing complex multi-agent interactions in real-time

Accomplishments that we're proud of

  • ⏱️ Full evaluation in under 1 minute
  • 📊 72 evaluation techniques across 6 domains
  • 🎨 Beautiful 3D visualization of agent orchestration
  • 📄 PDF export with detailed feedback

What we learned

  • LangGraph is powerful for multi-agent orchestration
  • Gemini 3's 1M context is a game-changer for code analysis
  • The sommelier metaphor makes AI evaluation approachable

What's next for Somm.dev

  • GitHub Action integration
  • VS Code extension
  • Custom evaluation "blends"
  • API for CI/CD pipelines

Built With

Share this project:

Updates