LangGraph for beginners : Agentic Workflows in simple steps

Discover Langgraph, built on lang chain, to orchestrate AI agents with graph-based workflows, node and edge design, memory, and conditional routing, enabling tool calls, retrieval, and streaming outputs.

Build a simple Langgraph workflow by defining a hello world state with a message key and adding hello and bye nodes, then invoke the compiled graph with Bharath.

Set up the LangGraph skeleton project by downloading the zip from the lecture resources, creating a new Python project with a virtual environment, and installing dependencies from requirements.txt.

Define a typed dict state and build hello and bye nodes in a lang graph workflow. Compile and run the graph with an initial state to see updates.

Define an entry point for a LangGraph workflow by setting a start edge to hello, ensuring a start point. Use graph.setEntryPoint or a start node to establish the entry.

Learn to visually represent your workflow by compiling a state graph, rendering a mermaid png, and displaying the runnable graph from start to hello to bye.

Explore state validation in Langgraph by passing at least one key from the state to the invoke method; see how missing inputs raise a key error.

Learn to validate the langgraph state using the Pydantic library by turning the state into a base model, applying min_length and max_length validations, and handling required and optional fields.

Explore LangGraph's asynchronous workflow execution with ainvoke, by converting nodes to async, using runnable to start the graph, and awaiting completion in an async main using asyncio.

Learn how an asynchronous workflow operates by turning two nodes into async, simulating with await sleep, and triggering api call, database call, and rag call, invoking the graph asynchronously.

Enable streaming in LangGraph, pass inputs to the stream method, and choose modes such as values, updates, messages, custom, and debug to stream results and improve workflow responsiveness.

Streaming in action demonstrates lang graph modes: values, updates, custom, messages, and debug. The demo uses a two-node graph with a stream writer to emit real-time chunks and state changes.

Learn conditional routing in Langgraph for customer service by priority, directing messages to urgent or standard nodes with a routing function.

Learn conditional routing with a two-node graph, urgent and standard, and use the categorize request logic to route messages by urgency or priority.

In LangGraph, define conditional edges with a routing function and a mapping dictionary to translate non-name outputs (like high or low) into node names such as urgent or standard.

Explore how reducer functions manage a LangGraph chatbot state, appending messages and discounts across nodes, and enable a 20% discount in the sales flow.

Learn to apply reducer functions to LangGraph state keys using the add function from the operator package and annotated from the typing library to append messages.

Learn how to use the inbuilt messages state class in Langgraph to manage a list of messages with a built-in reducer, simplifying state handling and avoiding extra fields.

Discover how LangGraph enables agentic workflows by binding tools to an llm and invoking them via the tool node. This session demonstrates a Get Restaurant Recommendations demo in Munich.

Explore tool calling with an llm by binding the restaurant recommendations tool, invoking the llm with a request to recommend restaurants in Munich, and extracting the tool call for execution.

Learn to use tool node in LangGraph by configuring available tools, constructing an ai message, and invoking tool calls to obtain responses from tools.

Build an agentic workflow by integrating a tool node with LLM, binding tools like Get Restaurant Recommendations, and routing tool calls through a conditional loop to the LLM for results.

In LangGraph for beginners, use ToolNode with LLM to bind tools, create a tool node, and route model calls via conditional edges for restaurant booking in Munich.

Learn why memory is essential in LangGraph workflows by re-invoking the graph with a remembered context to book a restaurant, compare responses, and reduce hallucinations.

Explore how Langgraph uses memory by passing state across nodes, storing with memory saver and unique thread id, and retrieving to continue, with permanent storage like PostgreSQL introduced later.

Learn to implement memory in a LangGraph workflow by creating a memory saver, setting a check pointer, and passing a config object to two invocations to retain session state.

Learn how Rag retrieval augmented generation powers an agentic workflow for a real time summarizer. Pull headlines from web sources, compute OpenAI embeddings, and store in chroma db for summarization.

Implement agentic RAG by loading data from URLs into a Chroma DB vector store, then retrieve and summarize current affairs with LangChain, using a prompt, model, and output parser.

Discover human in the loop patterns where an AI agent pauses for accept or reject decisions, reviews options, handles uncertainty, and edits state before proceeding.

Implement human in the loop with langgraph, using an interrupt and a command object to move through generate code, human review, and create test nodes.

Explore implementing human in the loop using LangGraph's interrupt function, handling interrupts, printing generated code, prompting user input, and resuming the graph to generate tests.

Pass additional values to an interrupt through a dictionary, then retrieve the data from the LangGraph task state and interrupt, using the API to extract the question for user input.