home / projects / PatchPilot: Multi-Agent Code Review System

PatchPilot: Multi-Agent Code Review System_

AI systems · Oct 2025 · code · by Nithin Jambula

A code review system where separate AI agents check every GitHub pull request for security, code quality and logic issues. A decision agent then approves or blocks the change based on its risk. The flow is a LangGraph state graph with conditional routing, so each pull request only goes through the checks it needs. Long-term memory in a ChromaDB vector store lets reviews learn from past issues. It also suggests patches for common problems and can switch between Gemini, Groq and Hugging Face models.

Built with

LangGraph · LangChain · ChromaDB · Gemini · Groq · GitHub API

More ai systems