LangGraph Tutorial: A Step-by-Step Guide to Creating a Text Analysis Pipeline

Estimated reading time: 5 minutes

Introduction to LangGraph

LangGraph is a powerful framework by LangChain designed for creating stateful, multi-actor applications with LLMs. It provides the structure and tools needed to build sophisticated AI agents through a graph-based approach.

Think of LangGraph as an architect’s drafting table – it gives us the tools to design how our agent will think and act. Just as an architect draws blueprints showing how different rooms connect and how people will flow through a building, LangGraph lets us design how different capabilities will connect and how information will flow through our agent.

Key Features:

State Management: Maintain persistent state across interactions
Flexible Routing: Define complex flows between components
Persistence: Save and resume workflows
Visualization: See and understand your agent’s structure

In this tutorial, we’ll demonstrate LangGraph by building a multi-step text analysis pipeline that processes text through three stages:

Text Classification: Categorize input text into predefined categories

Entity Extraction: Identify key entities from the text

Text Summarization: Generate a concise summary of the input text

This pipeline showcases how LangGraph can be used to create a modular, extensible workflow for natural language processing tasks.

Setting Up Our Environment

Before diving into the code, let’s set up our development environment.

Installation

# Install required packages
!pip install langgraph langchain langchain-openai python-dotenv

Setting Up API Keys

We’ll need an OpenAI API key to use their models. If you haven’t already, you can get one from https://platform.openai.com/signup.

Check out the Full Codes here

import os
from dotenv import load_dotenv

# Load environment variables from .env file (create this with your API key)
load_dotenv()

# Set OpenAI API key
os.environ[“OPENAI_API_KEY”] = os.getenv(‘OPENAI_API_KEY’)

Testing Our Setup

Let’s make sure our environment is working correctly by creating a simple test with the OpenAI model:

from langchain_openai import ChatOpenAI

# Initialize the ChatOpenAI instance
llm = ChatOpenAI(model=”gpt-4o-mini”)

# Test the setup
response = llm.invoke(“Hello! Are you working?”)
print(response.content)

Building Our Text Analysis Pipeline

Now let’s import the necessary packages for our LangGraph text analysis pipeline:

import os
from typing import TypedDict, List, Annotated
from langgraph.graph import StateGraph, END
from langchain.prompts import PromptTemplate
from langchain_openai import ChatOpenAI
from langchain.schema import HumanMessage
from langchain_core.runnables.graph import MermaidDrawMethod
from IPython.display import display, Image

Designing Our Agent’s Memory

Just as human intelligence requires memory, our agent needs a way to keep track of information. We create this using a TypedDict to define our state structure: Check out the Full Codes here

class State(TypedDict):
text: str
classification: str
entities: List[str]
summary: str

# Initialize our language model with temperature=0 for more deterministic outputs
llm = ChatOpenAI(model=”gpt-4o-mini”, temperature=0)

Creating Our Agent’s Core Capabilities

Now we’ll create the actual skills our agent will use. Each of these capabilities is implemented as a function that performs a specific type of analysis. Check out the Full Codes here

1. Classification Node

def classification_node(state: State):
”’Classify the text into one of the categories: News, Blog, Research, or Other”’
prompt = PromptTemplate(
input_variables=[“text”],
template=”Classify the following text into one of the categories: News, Blog, Research, or Other.\n\nText:{text}\n\nCategory:”
)
message = HumanMessage(content=prompt.format(text=state[“text”]))
classification = llm.invoke([message]).content.strip()
return {“classification”: classification}

2. Entity Extraction Node

def entity_extraction_node(state: State):
”’Extract all the entities (Person, Organization, Location) from the text”’
prompt = PromptTemplate(
input_variables=[“text”],
template=”Extract all the entities (Person, Organization, Location) from the following text. Provide the result as a comma-separated list.\n\nText:{text}\n\nEntities:”
)
message = HumanMessage(content=prompt.format(text=state[“text”]))
entities = llm.invoke([message]).content.strip().split(“, “)
return {“entities”: entities}

3. Summarization Node

def summarization_node(state: State):
”’Summarize the text in one short sentence”’
prompt = PromptTemplate(
input_variables=[“text”],
template=”Summarize the following text in one short sentence.\n\nText:{text}\n\nSummary:”
)
message = HumanMessage(content=prompt.format(text=state[“text”]))
summary = llm.invoke([message]).content.strip()
return {“summary”: summary}

Bringing It All Together

Now comes the most exciting part – connecting these capabilities into a coordinated system using LangGraph:

Check out the Full Codes here

# Create our StateGraph
workflow = StateGraph(State)

# Add nodes to the graph
workflow.add_node(“classification_node”, classification_node)
workflow.add_node(“entity_extraction”, entity_extraction_node)
workflow.add_node(“summarization”, summarization_node)

# Add edges to the graph
workflow.set_entry_point(“classification_node”) # Set the entry point of the graph
workflow.add_edge(“classification_node”, “entity_extraction”)
workflow.add_edge(“entity_extraction”, “summarization”)
workflow.add_edge(“summarization”, END)

# Compile the graph
app = workflow.compile()

Workflow Structure: Our pipeline follows this path:classification_node → entity_extraction → summarization → END

Testing Our Agent

Now that we’ve built our agent, let’s see how it performs with a real-world text example:

Check out the Full Codes here

sample_text = “”” OpenAI has announced the GPT-4 model, which is a large multimodal model that exhibits human-level performance on various professional benchmarks. It is developed to improve the alignment and safety of AI systems. Additionally, the model is designed to be more efficient and scalable than its predecessor, GPT-3. The GPT-4 model is expected to be released in the coming months and will be available to the public for research and development purposes. “””
state_input = {“text”: sample_text}
result = app.invoke(state_input)
print(“Classification:”, result[“classification”])
print(“\nEntities:”, result[“entities”])
print(“\nSummary:”, result[“summary”])
Classification: News Entities: [‘OpenAI’, ‘GPT-4’, ‘GPT-3’] Summary: OpenAI’s upcoming GPT-4 model is a multimodal AI that aims for human-level performance and improved safety, efficiency, and scalability compared to GPT-3.

Understanding the Power of Coordinated Processing

What makes this result particularly impressive isn’t just the individual outputs – it’s how each step builds on the others to create a complete understanding of the text.

The classification provides context that helps frame our understanding of the text type
The entity extraction identifies important names and concepts
The summarization distills the essence of the document

This mirrors human reading comprehension, where we naturally form an understanding of what kind of text it is, note important names and concepts, and form a mental summary – all while maintaining the relationships between these different aspects of understanding.

Try with Your Own Text

Now let’s try our pipeline with another text sample:

Check out the Full Codes here

# Replace this with your own text to analyze your_text = “”” The recent advancements in quantum computing have opened new possibilities for cryptography and data security. Researchers at MIT and Google have demonstrated quantum algorithms that could potentially break current encryption methods. However, they are also developing new quantum-resistant encryption techniques to protect data in the future. “””

# Process the text through our pipeline your_result = app.invoke({“text”: your_text}) print(“Classification:”, your_result[“classification”])

print(“\nEntities:”, your_result[“entities”])
print(“\nSummary:”, your_result[“summary”])

Classification: Research Entities: [‘MIT’, ‘Google’] Summary: Recent advancements in quantum computing may threaten current encryption methods while also prompting the development of new quantum-resistant techniques.

Adding More Capabilities (Advanced)

One of the powerful aspects of LangGraph is how easily we can extend our agent with new capabilities. Let’s add a sentiment analysis node to our pipeline:

Check out the Full Codes here

# First, let’s update our State to include sentiment
class EnhancedState(TypedDict):
text: str
classification: str
entities: List[str]
summary: str
sentiment: str

# Create our sentiment analysis node
def sentiment_node(state: EnhancedState):
”’Analyze the sentiment of the text: Positive, Negative, or Neutral”’
prompt = PromptTemplate(
input_variables=[“text”],
template=”Analyze the sentiment of the following text. Is it Positive, Negative, or Neutral?\n\nText:{text}\n\nSentiment:”
)
message = HumanMessage(content=prompt.format(text=state[“text”]))
sentiment = llm.invoke([message]).content.strip()
return {“sentiment”: sentiment}

# Create a new workflow with the enhanced state
enhanced_workflow = StateGraph(EnhancedState)

# Add the existing nodes
enhanced_workflow.add_node(“classification_node”, classification_node)
enhanced_workflow.add_node(“entity_extraction”, entity_extraction_node)
enhanced_workflow.add_node(“summarization”, summarization_node)

# Add our new sentiment node
enhanced_workflow.add_node(“sentiment_analysis”, sentiment_node)

# Create a more complex workflow with branches
enhanced_workflow.set_entry_point(“classification_node”)
enhanced_workflow.add_edge(“classification_node”, “entity_extraction”)
enhanced_workflow.add_edge(“entity_extraction”, “summarization”)
enhanced_workflow.add_edge(“summarization”, “sentiment_analysis”)
enhanced_workflow.add_edge(“sentiment_analysis”, END)

# Compile the enhanced graph
enhanced_app = enhanced_workflow.compile()

Testing the Enhanced Agent

# Try the enhanced pipeline with the same text
enhanced_result = enhanced_app.invoke({“text”: sample_text})

print(“Classification:”, enhanced_result[“classification”])
print(“\nEntities:”, enhanced_result[“entities”])
print(“\nSummary:”, enhanced_result[“summary”])
print(“\nSentiment:”, enhanced_result[“sentiment”])

Classification: News

Entities: [‘OpenAI’, ‘GPT-4’, ‘GPT-3’]

Summary: OpenAI’s upcoming GPT-4 model is a multimodal AI that aims for human-level performance and improved safety, efficiency, and scalability compared to GPT-3.

Sentiment: The sentiment of the text is Positive. It highlights the advancements and improvements of the GPT-4 model, emphasizing its human-level performance, efficiency, scalability, and the positive implications for AI alignment and safety. The anticipation of its release for public use further contributes to the positive tone.

Adding Conditional Edges (Advanced Logic)

Why Conditional Edges?

So far, our graph has followed a fixed linear path: classification_node → entity_extraction → summarization → (sentiment)

But in real-world applications, we often want to run certain steps only if needed. For example:

Only extract entities if the text is a News or Research article
Skip summarization if the text is very short
Add custom processing for Blog posts

LangGraph makes this easy through conditional edges – logic gates that dynamically route execution based on data in the current state.

Check out the Full Codes here

Creating a Routing Function

# Route after classification
def route_after_classification(state: EnhancedState) -> str:
category = state[“classification”].lower() # returns: “news”, “blog”, “research”, “other”
return category in [“news”, “research”]

Define the Conditional Graph

from langgraph.graph import StateGraph, END

conditional_workflow = StateGraph(EnhancedState)

# Add nodes
conditional_workflow.add_node(“classification_node”, classification_node)
conditional_workflow.add_node(“entity_extraction”, entity_extraction_node)
conditional_workflow.add_node(“summarization”, summarization_node)
conditional_workflow.add_node(“sentiment_analysis”, sentiment_node)

# Set entry point
conditional_workflow.set_entry_point(“classification_node”)

# Add conditional edge
conditional_workflow.add_conditional_edges(“classification_node”, route_after_classification, path_map={
True: “entity_extraction”,
False: “summarization”
})

# Add remaining static edges
conditional_workflow.add_edge(“entity_extraction”, “summarization”)
conditional_workflow.add_edge(“summarization”, “sentiment_analysis”)
conditional_workflow.add_edge(“sentiment_analysis”, END)

# Compile
conditional_app = conditional_workflow.compile()

Testing the Conditional Pipeline

test_text = “””
OpenAI released the GPT-4 model with enhanced performance on academic and professional tasks. It’s seen as a major breakthrough in alignment and reasoning capabilities.
“””

result = conditional_app.invoke({“text”: test_text})

print(“Classification:”, result[“classification”])
print(“Entities:”, result.get(“entities”, “Skipped”))
print(“Summary:”, result[“summary”])
print(“Sentiment:”, result[“sentiment”])

Classification: News
Entities: [‘OpenAI’, ‘GPT-4’]
Summary: OpenAI’s GPT-4 model significantly improves performance in academic and professional tasks, marking a breakthrough in alignment and reasoning.
Sentiment: The sentiment of the text is Positive. It highlights the release of the GPT-4 model as a significant advancement, emphasizing its enhanced performance and breakthrough capabilities.

Check out the Full Codes here

Now try it with a Blog:

blog_text = “””
Here’s what I learned from a week of meditating in silence. No phones, no talking—just me, my breath, and some deep realizations.
“””

result = conditional_app.invoke({“text”: blog_text})

print(“Classification:”, result[“classification”])
print(“Entities:”, result.get(“entities”, “Skipped (not applicable)”))
print(“Summary:”, result[“summary”])
print(“Sentiment:”, result[“sentiment”])

Classification: Blog
Entities: Skipped (not applicable)
Summary: A week of silent meditation led to profound personal insights.
Sentiment: The sentiment of the text is Positive. The mention of “deep realizations” and the overall reflective nature of the experience suggests a beneficial and enlightening outcome from the meditation practice.

With conditional edges, our agent can now:

Make decisions based on context
Skip unnecessary steps
Run faster and cheaper
Behave more intelligently

Conclusion

In this tutorial, we’ve:

Explored LangGraph concepts and its graph-based approach

Built a text processing pipeline with classification, entity extraction, and summarization

Enhanced our pipeline with additional capabilities

Introduced conditional edges to dynamically control the flow based on classification results

Visualized our workflow

Tested our agent with real-world text examples

LangGraph provides a powerful framework for creating AI agents by modeling them as graphs of capabilities. This approach makes it easy to design, modify, and extend complex AI systems.

Next Steps

Add more nodes to extend your agent’s capabilities
Experiment with different LLMs and parameters
Explore LangGraph’s state persistence features for ongoing conversations

Check out the Full Codes here. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 100k+ ML SubReddit and Subscribe to our Newsletter.

You may also like NVIDIA’s Open Sourced Cosmos DiffusionRenderer [Check it now]

Nir Diamant is an AI researcher, algorithm developer, and specialist in GenAI, with over a decade of experience in AI research and algorithms. His open-source projects have gained millions of views, with over 500K monthly views and over 50K stars on GitHub, making him a leading voice in the AI community.

Through his work on GitHub and the DiamantAI newsletter, Nir has helped millions improve their AI skills with practical guides and tutorials.

Source link