PR Review Agent

AI-powered tool that reviews any GitHub pull request in seconds — scoring Security, Performance, Code Quality & Maintainability with line-level fix suggestions.

Comment

🚀 Inspiration

Every developer knows the pain — you submit a pull request and wait hours, sometimes days, for a code review. Senior engineers are overloaded. Junior developers get inconsistent feedback. Critical security bugs slip through because reviewers are tired or rushed.

I wanted to fix that. What if any developer could get an expert-level code review in seconds, on any PR, for free?

🔨 What I Built

PR Review Agent is a fully deployed, production-ready AI tool that reviews any public GitHub Pull Request instantly.

Paste a PR URL → get back:

  • An overall score out of 100
  • Individual scores for Security, Performance, Code Quality, and Maintainability
  • Line-level issue detection with specific fix suggestions
  • A clear verdict on whether the code is ready to merge

⚙️ How I Built It

Frontend: React + Vite + Tailwind CSS, deployed on Vercel

Backend: FastAPI (Python), deployed on Railway

AI: Llama 3.3 70B via Groq API — chosen for its speed, code reasoning ability, and free tier

Data: GitHub REST API to fetch full PR diffs programmatically

The backend fetches the raw diff from GitHub, structures it into a detailed prompt, sends it to Llama 3.3 70B, parses the structured JSON response, and returns scored results to the frontend in seconds.

🧗 Challenges I Faced

  • JSON parsing reliability — LLMs sometimes return malformed JSON. Built a regex-based fallback parser with trailing comma fixer to handle edge cases gracefully.

  • Large PR diffs — Some PRs have thousands of changed lines. Had to implement diff truncation logic to stay within token limits while preserving the most critical changes.

  • Error handling across the stack — Making sure every failure (bad URL, private repo, API timeout) returns a clear, human-readable message instead of a cryptic crash.

  • Deployment configuration — Getting FastAPI running correctly on Railway with the right Procfile, nixpacks config, and environment variables took several iterations.

📚 What I Learned

  • How to build a full production API pipeline around LLMs
  • Prompt engineering for structured, reliable JSON output
  • Deploying a full-stack app (Vercel + Railway) end to end
  • How to handle real-world edge cases in AI-powered applications

🌍 Real-World Impact

This tool saves senior engineers hours per week. It gives junior developers instant, consistent feedback on every PR. It catches security vulnerabilities before they reach production. It democratizes expert code review for solo developers and small teams who don't have senior engineers available.

🔗 Try It Live

👉 https://pr-review-agent-swart.vercel.app/

Paste any public GitHub PR URL and see it in action.

https://github.com/vivekyarra/pr-review-agent

Built With

Share this project:

Updates