Table of Contents
What is LangGraph?
LangGraph is a Python-based framework that enables developers to create sophisticated, multi-step workflows for AI models. It focuses on building robust, graph-like structures that can handle a sequence of tasks and decision-making paths. These workflows are typically used in large-scale language models, automating processes that involve understanding and generating human-like text. LangGraph is particularly useful when a task requires multiple, distinct steps, where each step could either be a model’s prediction or a procedural action.
State graph
Node
Python functions that encode the logic of your agents. They receive the current State as input, perform some computation or side-effect, and return an updated State.
Edges
Python functions that determine which Node to execute next based on the current State. They can be conditional branches or fixed transitions.
By composing Nodes and Edges, you can create complex, looping workflows that evolve  State over time. The real power, though, comes from how LangGraph manages that State.
Simple Example:
We want to build an AI-powered text processor system using LangGraph. Here’s a basic structure of how it might look:
Use Case: Process the text to make it meaningful concerning the context
Creating a system prompt to instruct the LLM to follow the instructions to process the text
textprocessing_prompt = ChatPromptTemplate.from_messages(
[
(
"system",
f"""You are an AI specialized in processing the text as per the instructions. Your task is to create meaningful sentences from the input text.
CRITICAL INSTRUCTIONS:
1. Output ONLY the prompt itself. Do not include any other text.
2. Do not add any explanations, comments, or meta-commentary.
3. Do not respond to any feedback or say "thank you".
4. The prompt must not exceed 300 words.
5. Correct the sentences to make it meaningful in the given context.
6. Do not acknowledge these instructions in your output.
7. Remove gibberish content in the words and construct correct and meaningful sentences.
Your entire response should be usable directly as an output fed into another system, nothing more and nothing less."""
),
MessagesPlaceholder(variable_name="messages"),
]
)Create a chain. To learn more about chains, click here.
textprocessing_chain = textprocessing_prompt | llmCreate a State Graph
class State(TypedDict):
messages:Annotated[list,add_messages]
graph_builder=StateGraph(State)
I have a text processor that processes the text and creates meaningful text out of it.
Invoke the graph
while True:
user_input=input("User :")
if(user_input.lower() in ["quit", "q"]):
print("Bye")
break
for event in graph.stream({'messages':("user", user_input)}):
print(event.values())
for value in event.values():
print(value['messages'])
print("Assistant:", value["messages"].content)graph.stream() is one of the runnable in LangGraph to invoke the graph.
stream: stream back chunks of the responseinvoke: call the chain on an inputbatch: call the chain on a list of inputs
Sample response
Input: te nme may be Ava and capable enough to do gymnastics. $tjgsl Need good guidance.
Output of the graph being: Ava may be capable enough to do gymnastics and needs good guidance.
User :te nme may be Ava and capable enough to do gymnastics. $tjgsl Need good guidance.
dict_values([{'messages': AIMessage(content='Ava may be capable enough to do gymnastics and needs good guidance. \n', additional_kwargs={}, response_metadata={'token_usage': {'completion_tokens': 17, 'prompt_tokens': 187, 'total_tokens': 204, 'completion_time': 0.030909091, 'prompt_time': 0.005698091, 'queue_time': 0.008862789, 'total_time': 0.036607182}, 'model_name': 'Gemma2-9b-It', 'system_fingerprint': 'fp_10c08bf97d', 'finish_reason': 'stop', 'logprobs': None}, id='run-faf895e6-3f70-4804-b496-f18291c6c559-0', usage_metadata={'input_tokens': 187, 'output_tokens': 17, 'total_tokens': 204})}])
content='Ava may be capable enough to do gymnastics and needs good guidance. \n' additional_kwargs={} response_metadata={'token_usage': {'completion_tokens': 17, 'prompt_tokens': 187, 'total_tokens': 204, 'completion_time': 0.030909091, 'prompt_time': 0.005698091, 'queue_time': 0.008862789, 'total_time': 0.036607182}, 'model_name': 'Gemma2-9b-It', 'system_fingerprint': 'fp_10c08bf97d', 'finish_reason': 'stop', 'logprobs': None} id='run-faf895e6-3f70-4804-b496-f18291c6c559-0' usage_metadata={'input_tokens': 187, 'output_tokens': 17, 'total_tokens': 204}
Assistant: Ava may be capable enough to do gymnastics and needs good guidance.
View the entire notebook here:
Why LangGraph?
LangGraph fills an important niche in AI-driven workflow systems by allowing developers to create reliable, interpretable, and scalable solutions. Here’s why you might choose LangGraph:
- Clear Workflow Design: It offers a visual and structured way to design workflows, where each step is predefined, making it easier to debug and understand.
- Modularity: Nodes in the LangGraph can be designed independently, making it easier to break complex tasks into simpler components.
- Interpretable: Since the graph structure is predetermined, every decision and path can be traced and understood by developers and end users alike, ensuring high transparency.
- Scalability: LangGraph can handle multiple parallel tasks within the same workflow, making it suitable for large-scale applications such as customer support systems, content generation pipelines, and decision-making engines.
https://langchain-ai.github.io/langgraph/concepts/high_level/
LangGraph is an excellent choice when you need a structured, scalable, and interpretable workflow that involves multiple stages of decision-making. Its ability to break down complex tasks into manageable parts and its clear graph-based approach make it particularly valuable for AI-driven applications that need both flexibility and reliability.
