• 

  CodexAir: AI-Powered Code Review Intelligence Platform

🚀 Inspiration

Modern code reviews are often time-consuming, subjective, and inconsistent across teams. We built CodexAir to empower developers and teams with real-time, AI-powered insights—enhancing code quality, detecting vulnerabilities early, and promoting engineering excellence without slowing development cycles.

🧠 What it does

CodexAir automates and enriches code reviews on GitHub with AI and static analysis. It provides:

• ✅ Automated code quality scoring
  
• 🔐 Security vulnerability detection with CVE/CWE mapping
  
• 🚀 Performance and refactoring suggestions
  
• 🧬 Semantic duplicate code detection via vector search
  
• 📊 Dashboards with team metrics, quality trends, and PR comparisons
  
• 💬 Inline suggestions and auto-fix patches through GitHub comments
  

🏗️ How we built it

CodexAir is a full-stack, serverless platform built with:

• Frontend & Backend: Next.js 14 (App Router) with API routes
  
• Database: MongoDB Atlas + Vector Search for embeddings & metadata
  
• AI Analysis: Google Vertex AI (Gemini Pro) for code scoring & embeddings
  
• Authentication: NextAuth.js + GitHub OAuth
  
• PR Sync & Comments: GitHub API (Octokit)
  
• UI: ShadCN UI, Radix UI, Recharts
  
• Caching: Redis for GitHub API rate limits
  
• Deployment: Vercel with CI/CD via GitHub Actions
  

🧗 Challenges we ran into

• Efficiently handling large diffs and multi-file PRs
  
• Tuning AI prompts for consistent and accurate suggestions
  
• Managing GitHub rate limits with caching and retries
  
• Aligning vector similarity logic to real-world developer expectations
  
• Embedding non-intrusive inline suggestions into GitHub workflows
  

🏆 Accomplishments that we're proud of

• Created a complete AI-powered code review loop
  
• Integrated CWE/CVE scanning into live PR analysis
  
• Built semantic vector search to detect duplicate code across PRs
  
• Designed a production-ready dashboard with live metrics
  
• Deployed seamlessly on Vercel with CI/CD pipelines
  

📚 What we learned

• How to combine LLM reasoning with static analysis
  
• The capabilities and limitations of AI embeddings for code
  
• The importance of developer trust in AI-assisted tools
  
• Advanced usage of MongoDB Vector Search, Vertex AI, and GitHub APIs
  

🔮 What's next for CodexAir

• 🌐 Multi-language support (currently JavaScript/TypeScript)
  
• 🔥 Codebase-wide heatmaps and technical debt tracking
  
• 🛠️ Custom rule definitions for enterprise use
  
• 📈 Team analytics + Slack/Discord alerts
  
• 🧾 Compliance-ready exports (SOC2, ISO27001, etc.)
  
• 🧩 Plugin marketplace for stack-specific AI review models
  

🛠 Built with

Next.js 14, TypeScript, MongoDB Atlas, MongoDB Vector Search, Google Vertex AI (Gemini Pro), Redis, GitHub OAuth, GitHub API (Octokit), ShadCN UI, Radix UI, Recharts, NextAuth.js, Vercel, GitHub Actions

Built With

• genimi
• github
• github-api-(octokit)
• github-oauth
• mongodb-atlas
• mongodb-vector-search
• next.js-14
• nextauth.js
• radix-ui
• recharts
• redis
• shadcn-ui
• typescript
• vercel