Course Overview
This course teaches you to build AI agents in Python by creating six different AI agent projects from scratch. Each module introduces a concept and applies it in a working project, covering structured output, tool calling, MCP, RAG, LangGraph, and an agent that audits its own work. You write every line of code yourself, gaining a deep understanding of each component.
The course runs entirely on local models with Ollama, with optional cloud API integration. It is designed for developers with basic programming knowledge in any language, with Python taught as needed.
Key Takeaways
- Build six working AI agents in Python from scratch, understanding every line of code.
- Create a complete RAG pipeline including chunking, embeddings, ChromaDB, retrieval, and memory management.
- Master LangGraph features such as typed state, parallel branches, conditional routing, cycles, checkpointers, and human approval.
- Develop your own MCP server and client, integrate with Claude Desktop, and use community MCP servers.
- Build an agentic RAG that self-routes queries, rewrites bad queries, retries retrievals, filters by metadata, and audits documents.
- Understand tool calling and ReAct by implementing loops manually before using LangChain.
- Debug agents using LangSmith traces with real-time bug fixing demonstrations.
Prerequisites
- Basic programming knowledge in any language (Java, C#, Go, JavaScript, PHP, etc.); Python is not required beforehand.
- A laptop with 16 GB RAM recommended (8 GB acceptable for cloud API use).
- Python 3.13, uv, and Git installed (covered in setup lecture).
Target Learners
- Backend, full-stack, and mobile developers transitioning into AI engineering.
- Developers familiar with languages other than Python who want to build AI agents.
- CS students and self-taught coders seeking a portfolio of working AI agents.
- Individuals who have used AI tools for agents but want to understand and fix their code.
- Learners who found other agent courses too fast-paced or reliant on pre-written code.
Every AI Course Starts the Same Way. This One Doesn't.
5:23
How to Follow Along: Your Project, My Reference Repo
3:35
What Is an AI Agent?
5:05
Setup for both Mac and Windows
6:03
Your First LLM Call from Python
4:54
Udemy Rating
1:01
What You'll Build: An Email-to-CRM Parser
3:22
Inside an LLM Call: Messages, Roles, and Temperature
3:56
LiteLLM: One Interface for Every LLM Provider
4:35
Your First API Call and the 4 Errors You'll Hit
5:42
Tokens, Pricing, and Choosing the Right Model
5:12
API Keys, .env, and the One-Line Provider Switch
8:54
Why Free-Text LLM Output Breaks Your Code
4:44
Pydantic Basics: Schemas as Python Classes
6:00
Enums and Constraints: Stop the LLM Inventing Values
3:55
Custom Validators and Reading Pydantic Errors
3:27
Instructor: Typed, Validated Objects, Auto-Retried
9:11
Streamlit UI: Run Your Parser from the Browser
9:58
Quiz — Email-to-CRM Parser
Mentor the New Intern: Why We Validate LLM Output
What You'll Build: A Stock Research Agent
2:56
The ReAct Loop: Reason, Act, Observe, Repeat
4:28
Function Calling by Hand: Write the Loop Yourself
12:46
Add Tool Calling to Your Wrapper (and Python Generics)
10:43
Your First Agent Tool: Live Stock Prices with @tool
8:13
LangChain create_agent: Replace Your Hand-Rolled Loop
6:11
A Second Tool: Web Search and Tool Routing
10:22
Structured Output from an Agent: Gather, Then Format
11:14
LangSmith Tracing: See Every Step Your Agent Takes
8:31
Reading a LangSmith Trace to Debug an Agent
8:33
Streamlit Chat UI for Your Agent
7:31
Quiz — The Logic Loop (Live Research Assistant)
What You'll Build: An On-Call Assistant, Powered by MCP
7:42
MCP Architecture: Servers, Clients, Hosts, JSON-RPC
4:34
Build Your First MCP Server with FastMCP
9:05
A Real MCP Server: Tools, Pydantic, and Resources
11:20
MCP Transports: stdio vs HTTP
6:55
Build an MCP Client from Scratch, No Framework
10:44
Plug MCP Tools into a LangChain Agent
14:21
Connect Your MCP Server to Claude Desktop
7:06
Use a Community MCP Server: Filesystem
14:00
Use a SaaS MCP Server: Notion, with Token Auth
14:43
Project: Meeting-Prep Agent with Calendar and Notion
20:53
Quiz — The Connected Agent (Engineering Onboarding Assistant)
What You'll Build: A Document Q&A Agent with Memory
3:00
What Is RAG, and Why Your Agent Needs It
4:06
Ask an LLM About a Document It's Never Seen
7:10
Embeddings Explained: Turning Text into Vectors
8:58
Chunking and Indexing Documents into ChromaDB
19:36
Similarity Search: Retrieving the Right Chunks
10:27
The Full RAG Pipeline: Retrieve, Augment, Answer
15:30
Chunk Size, Metadata, and the 4 Ways RAG Breaks
19:24
Why Your Agent Forgets: The Follow-Up It Can't Answer
3:59
Short-Term Memory: A Sliding Window of Turns
8:35
Long-Term Memory: Summaries in a Vector Store
13:31
Fusing Document Chunks and Memory into One Prompt
19:49
Quiz — The Deep Reader, Part 1 (Medical Policy Q&A Assistant)
What You'll Build: An Analyst That Reads a Pitch Deck
6:31
The LangGraph Mental Model: Nodes, Edges, State
7:09
Your First StateGraph (and TypedDict State)
19:31
Subgraphs: One Factory, Three Branches
23:08
Parallel Branches: Fan-Out, Fan-In, and Reducers
10:43
Conditional Routing: Let the Graph Choose Its Path
14:09
Closing the Graph with a Structured Report
15:12
Checkpointers and thread_id: Pause, Resume, Replay
9:07
Human-in-the-Loop: Interrupt, Inject, Resume
13:53
Multimodal Vision: Reading Slides with an LLM
23:37
LangGraph Studio: Visually Debug and Replay a Graph
16:08
When LangGraph Is Overkill, and What It Costs
7:54
Exercise: Add a Fifth Parallel Branch
7:48
Quiz — LangGraph Deep Dive (Investment Analyst)
What You'll Build: An Agent That Audits Any Document
5:30
Why Naive RAG Fails, and What Agentic RAG Adds
5:46
Build a 4-Route Query Router
15:29
Query Rewriting and the Self-Correction Loop
12:24
Metadata Filtering: Narrow Retrieval Before Search
18:30
Full-Document Audit with Structured Output
20:28
One Agent, Any Document: Pluggable Doc Profiles
13:59
Wiring the Full Agentic RAG Graph
19:57
First Run: It's Broken. Read the Trace, Fix the Bugs
25:53
Hand-Rolled to Framework: The Vector Store
13:20
Hand-Rolled to Framework: Retrieval and Documents
11:40
Hand-Rolled to Framework: Checkpointer Memory
11:17
Ship It: A Chainlit Chat UI for the Deep Reader
10:24
System Design Interview: A Pitch Deck Agent
Access Files
No pi required.
Free
Course files
1 file package
-
Full Pack
Discussions are closed
Comments are currently disabled for this course.
