Build AI Agents in Python: LangChain, LangGraph, MCP & RAG

This course teaches you to build AI agents in Python by building six of AI agents. Each module takes one concept, explains it, and then uses it in a working project, so that by the end you know every piece an agent needs: structured output, tool calling, MCP, RAG, LangGraph, and an agent that checks its own work.

What you will build
Six projects, one per module, and each one runs by the end of its module:

• An email-to-CRM parser that turns messy emails into validated records, built with LiteLLM, Pydantic and Instructor, with a Streamlit UI on top.

• A stock research agent that calls a live price tool and a web search tool in a ReAct loop. You write the loop by hand first, then replace it with LangChain, and trace every step in LangSmith.

• Your own MCP server and MCP client, plugged into a LangChain agent and into Claude Desktop, and then a meeting-prep agent that reads a calendar and Notion through MCP.

• A document Q&A agent with RAG built by hand: chunking, embeddings, ChromaDB, similarity search, and short-term and long-term memory so that it can answer follow-up questions.

• A pitch deck analyst in LangGraph with typed state, parallel branches, conditional routing, checkpointers, a pause for human approval, vision on the slides, and debugging in LangGraph Studio.

• An agentic RAG agent that routes its own questions, rewrites bad queries, retries failed retrieval, audits a whole document, and ships with a Chainlit chat UI.

How the course works

• You build every line yourself, with GitHub Copilot for completions, the way a working developer does. Nothing is pasted in from a finished file. Every new thing, from Pydantic to decorators to async to graphs, is explained the moment it first appears and before it goes into your project, so that when something breaks you know why.

• Everything runs for free on a local model with Ollama. Switching to OpenAI or Claude is one line in a .env file, because the code is written against LiteLLM from the first lesson.

• You build in your own project, not a clone of mine. My reference repo has a tag for the end of every code lesson, so if you fall behind you put your files next to mine and fix the difference.

Who this course is for

• Developers in any language who want to build AI agents and have put it off because everything is in Python. You do not need Python. Basic knowledge of any programming language is enough. I came from the enterprise backend world myself and learned Python by building agents.

• Students and self-taught coders who want a portfolio of working agents, not just a certificate.

What this course is not

It is not a machine learning course. We do not train models and there is no math. It is not a Python course either. Python is taught as it is needed, one piece at a time.

Requirements

A laptop with 16 GB of RAM to run local models comfortably, or 8 GB if you use a cloud API. Python 3.13, uv and Git are installed together in the setup lecture.