Back to Projects
Agentic AIadvanced

Advanced AI GitHub PR Code Reviewer

This project is an intelligent, event-driven GitHub Pull Request review system designed to automatically trigger on PR events using GitHub Webhooks. It processes code diffs through a multi-agent AI pipeline built with LangGraph and powered by OpenAI GPT-4o-mini. The system runs four parallel review agents—Static Analysis, Security (aligned with OWASP Top 10), Architecture, and Code Style—which collaboratively analyze the code and merge their findings into structured, actionable feedback. The backend architecture leverages FastAPI-based microservices, with Celery and Redis handling asynchronous task queuing, while PostgreSQL (using asyncpg and SQLAlchemy) manages data persistence. Additionally, a self-improving Learner service continuously refines style patterns based on merged pull requests, enabling personalized and evolving code reviews over time. For observability, the system integrates Prometheus and Grafana for metrics monitoring, along with Langfuse for detailed LLM tracing and performance insights.

34 lectures

What You Will Learn

Students will learn how to design and orchestrate parallel AI agents using LangGraph’s StateGraph. This includes running multiple specialized agents—Static Analysis, Security, Architecture, and Code Style—simultaneously and combining their outputs into unified, structured feedback for code reviews.
Students will learn how to build production-grade, event-driven microservices using FastAPI. The system will handle real-world GitHub Webhook events, process tasks asynchronously using Celery with Redis as the message broker, and store results efficiently in PostgreSQL.
Students will learn how to implement observability for LLM-based systems using Langfuse for tracing and monitoring, along with Prometheus and Grafana for system metrics. They will also build a feedback loop through a Learner service that continuously improves code review quality by learning patterns from merged pull requests.

System Architecture

Advanced AI GitHub PR Code Reviewer Architecture Diagram

High-level architecture overview of the Advanced AI GitHub PR Code Reviewer .

What You'll Build

  • We will build a LangGraph-powered code review engine that automatically fetches GitHub pull request diffs and processes them using four parallel AI agents: Static Analysis, Security Scanner, Architecture Reviewer, and Code Style Checker. Powered by OpenAI GPT-4o-mini, these agents will collaboratively generate structured, actionable feedback for every pull request.
  • We will build a scalable backend system composed of five FastAPI-based microservices: Gateway, Webhook, Orchestrator, Reviewer, and Learner. These services will communicate asynchronously using Celery with Redis as the message broker, while PostgreSQL will handle data persistence. The system will also implement JWT-based GitHub App authentication to ensure secure API access.
  • We will develop a Learner service that continuously analyzes merged pull requests to extract evolving code style patterns, enabling personalized and improving code reviews over time. Additionally, we will integrate a complete observability stack using Prometheus for metrics collection, Grafana for visualization, and Langfuse for LLM tracing—ensuring full visibility and continuous system improvement.

Project Instructor

Sudhanshu

Sudhanshu

4+ years exp
LinkedIn
Advanced AI GitHub PR Code Reviewer
Premium
One Subscription. 40+ Projects. Unlimited Access.
AccessMobile & Web