LLM App Development with LangChain
Chains, retrieval, agents and deployment with the most-used LLM framework.
Taught by Rohan Maharjan · ML Engineer
What you'll learn
- Set up LangChain 1.x with a chat model, prompt templates and output parsers composed as runnables
- Return validated structured output from a model using Pydantic schemas
- Build a RAG pipeline with loaders, splitters, embeddings, a vector store and a retrieval chain
- Create a tool-calling agent with LangGraph state, nodes, edges and persistence
- Stream tokens and intermediate events to a client with sync and async APIs
- Trace, evaluate and deploy a LangChain app with LangSmith and FastAPI
About this course
LLM App Development with LangChain takes you from a first chat model call to a deployed application using LangChain 1.x, LangGraph and LangSmith. It is for Python developers who want to build production LLM apps rather than notebooks: you should be comfortable with functions, classes, virtual environments and basic async, and have used a chat model API before. Across eight hands-on lessons you will write prompt templates and chains, get typed structured output with Pydantic, build a retrieval-augmented generation pipeline over your own documents, create a tool-using agent with LangGraph, stream tokens to a user interface, evaluate and trace your app with LangSmith, and deploy it behind a FastAPI endpoint. Every lesson includes runnable code against an OpenAI-compatible model, and every pattern transfers to other providers by changing one line. You finish with a complete, observable question-answering service and the judgement to know when LangChain helps and when plain Python is enough.
Course content
8 lessons- 1The LangChain Ecosystem in 2026What LangChain, LangGraph and LangSmith each do, how the packages fit together, and your first model call.Preview
- 2Prompts, Models and ChainsCompose prompt templates, models and parsers into runnables with the pipe operator, batching and async.
- 3Structured Output with PydanticGet typed, validated objects from a model with with_structured_output and handle failures gracefully.
- 4Retrieval-Augmented GenerationLoad, split, embed and index documents, then answer questions with cited context in a retrieval chain.
- 5Agents with LangGraphDefine tools, build a tool-calling agent, then open the graph to add custom state, nodes, edges and persistence.
- 6Streaming and AsyncStream tokens and intermediate events from chains and graphs, and serve them from async code without blocking.
- 7Evaluation and Tracing with LangSmithTrace every run, build a dataset of examples, write evaluators and compare versions before you change production.
- 8Deploying a LangChain AppServe the app with FastAPI and streaming, manage configuration and secrets, add limits, and containerise it.