LangGraph Code Review Workshop

A hands-on tutorial for building a multi-agent code review system using LangGraph.

Workshop Overview

This workshop teaches you to build a LangGraph application that:

1. Generates code based on requirements
2. Reviews the generated code
3. Refactors the code based on review feedback

Prerequisites

• Python 3.12+
• OpenAI API key
• Basic understanding of Python and AI concepts

Environment Manager Installation (Skip this step if you already use Conda or Venv)

Checkpoint: Install environment manager

• Mac/Linux:
  • Install Conda (or can use venv if preferred)
• Windows:
  • Skip this step. You can just use global pip package manager.

What you should have: An environment manager

Project Setup

• Create a new directory called langgraph-task-1 with a file called main.py inside it
• Conda users (Mac/Linux), use conda create -n langgraph-task-1, then conda activate langgraph-task-1 inside the directory

  • Venv users, setup an environment for this project as you normally would

What you should have: An environment to work in

Package Installation

Inside your project directory, run:

pip install langchain-openai langgraph

Step-by-Step Tutorial

Step 1: Import Dependencies

Checkpoint: Basic imports

Add the required imports to main.py:

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langgraph.graph import StateGraph, END, START

What you should have: A file with the basic LangGraph and LangChain imports.

---

Step 2: Setup the LLM

Checkpoint: LLM configuration

Initialize the OpenAI model:

llm = ChatOpenAI(model="gpt-4", api_key="YOUR API KEY HERE")

Note: Replace "YOUR API KEY HERE" with your actual OpenAI API key.

What you should have: An LLM instance ready to use in your agents.

---

Step 3: Create Agent Prompts

Checkpoint: Prompt templates

Define prompts for each agent role:

coder_prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a Coder. Write Python code based on the given requirements."),
    ("human", "{input}")
])

reviewer_prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a Reviewer. Review the given code and suggest improvements."),
    ("human", "{code}")
])

refactorer_prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a Refactorer. Implement the suggested improvements in the code."),
    ("human", "Code:\n{code}\n\nReview:\n{review}")
])

What you should have: Three distinct prompt templates for coder, reviewer, and refactorer roles.

---

Step 4: Implement Agent Functions

Checkpoint: Agent implementations

Create the three agent functions:

def coder_agent(state):
    response = llm.invoke(coder_prompt.format_messages(input=state["input"]))
    state["code"] = response.content
    return state

def reviewer_agent(state):
    response = llm.invoke(reviewer_prompt.format_messages(code=state["code"]))
    state["review"] = response.content
    return state

def refactorer_agent(state):
    response = llm.invoke(refactorer_prompt.format_messages(
        code=state["code"], review=state["review"]))
    state["refactored_code"] = response.content
    return state

What you should have: Three functions that each take the state, call the LLM with appropriate prompts, and update the state.

---

Step 5: Build the Graph

Checkpoint: Graph construction

Create the LangGraph workflow:

builder = StateGraph(dict)
builder.add_node("coder", coder_agent)
builder.add_node("reviewer", reviewer_agent)
builder.add_node("refactorer", refactorer_agent)

builder.add_edge(START, "coder")
builder.add_edge("coder", "reviewer")
builder.add_edge("reviewer", "refactorer")
builder.add_edge("refactorer", END)

graph = builder.compile()

What you should have: A compiled graph that defines the workflow: START → coder → reviewer → refactorer → END.

---

Step 6: Test the System

Checkpoint: Working application

Add the example usage:

task = "Write a function that checks if a string is a palindrome"
initial_state = {"input": task}
final_state = graph.invoke(initial_state)

print("======================================")
print("INITIAL CODE")
print("======================================")
print("\nInitial Code:\n", final_state["code"])

print("\n======================================")
print("CODE REVIEW")
print("======================================")
print("\nReview Feedback:\n", final_state["review"])

print("\n======================================")
print("REFACTORED CODE")
print("======================================")
print("\nRefactored Code:\n", final_state["refactored_code"])

What you should have: A complete working application that generates, reviews, and refactors code.

---

Running the Application

python main.py

Common Issues & Debugging

Issue: ImportError

Solution: Ensure all packages are installed: pip install langchain-openai langgraph

Issue: API Key Error

Solution: Set your OpenAI API key in the code or as an environment variable

Issue: Graph doesn't execute

Solution: Check that all edges are properly connected and the graph is compiled

Issue: Agent functions not working

Solution: Verify each agent function returns the updated state

Next Steps

• Experiment with different prompts
• Add error handling
• Implement conditional logic in the graph
• Add human-in-the-loop interactions
• Extend with additional agents (e.g., tester, documentation writer)

Support

If you're stuck at any step, reference the step number and checkpoint description for targeted help.