本课程教你用纯Python从零构建生产级AI Agent，手写工具调用循环、RAG检索算法与MCP协议服务器客户端，并通过FastAPI和Docker安全部署，全程可在8GB笔记本本地免费运行。
原始标题：AI Agents with LangGraph & MCP: RAG, Tool Calling, Multi-Age
这门课程教你不依赖 LangChain 等任何第三方框架，纯手写 Python 代码从零构建一个生产级别的 AI Agent。你将亲手实现大模型工具调用循环、手写 RAG 检索算法、编写最新 MCP 协议的服务器与客户端，并最终通过 FastAPI 和 Docker 将其安全部署上线，整个开发流程均可在 8GB 内存的笔记本上完全本地、免费运行。
Last updated 10/2026
Created by Vaibhav Kumar Singh
MP4 | Video: h264, 1920×1080 | Audio: AAC, 44.1 KHz, 2 Ch
Level: All Levels | Genre: eLearning | Language: English + subtitle | Duration: 56 Lectures ( 2h 6m ) | Size: 1.9 GB
Code the agent loop yourself in Python, connect it to tools and documents, then test, secure and deploy it.
What you’ll learn
⚡ Build a working AI agent loop from scratch in Python: tool calling, observations, step limits and error recovery
⚡ Get reliable structured output from LLMs with JSON schemas, Pydantic validation and repair loops
⚡ Build search from scratch (BM25, embeddings, vector search, hybrid search) and use it for retrieval augmented generation (RAG)
⚡ Connect agents to tools and data with the Model Context Protocol (MCP), writing both a server and a client
⚡ Design memory, planning and multi-agent systems, and know when a single agent is the better choice
⚡ Evaluate agents with test sets, LLM-as-judge and tracing, and defend them against prompt injection
⚡ Run everything locally on a laptop for free, or switch to Claude with one setting, while tracking tokens and cost
⚡ Deploy an agent with FastAPI, Docker, a web UI and monitoring
Requirements
❗ Basic Python: functions, classes, dictionaries and lists
❗ A laptop with about 8 GB of RAM. No GPU and no paid API key required
❗ Optional: a Claude API key, if you want faster runs
Description
This course contains the use of artificial intelligence.
AI agents are the fastest-moving area in software. Most tutorials teach you a framework’s functions. This course teaches you how agents actually work, by building one from an empty folder.
You will write the agent loop by hand: the model decides which tool to call, your code runs it, and the result goes back to the model until the task is done. Then you will add each capability that real agents need, one section at a time.
**Everything runs on your laptop for free.** The default model is Qwen3 4B, an open model that runs offline through llama.cpp. No GPU and no API key needed. If you have a Claude API key, switch with one setting and every example runs faster, while the course shows you the exact token cost.
**What you will build**
✅ A small agent library, agentkit, with one interface for local and hosted models
✅ Reliable structured output with JSON schemas, Pydantic validation and repair loops
✅ Search from scratch: BM25, embeddings, vector search and hybrid search
✅ Retrieval augmented generation (RAG) with chunking, citations and reranking
✅ Tool calling and a ReAct agent loop, including error recovery and step limits
✅ An MCP server and client, using the Model Context Protocol adopted across the industry
✅ Memory, planning and multi-agent systems, plus the same design in LangGraph
✅ Evaluations, tracing and regression tests for agents
✅ Defences against prompt injection, human approval for risky actions, and cost control
✅ A deployed finance operations agent with FastAPI, Docker, a web UI and monitoring
**Real results, including failures.** Every output in the videos comes from real runs. When the model makes a mistake, you see it, and you learn how to fix the system around it. This is the skill that separates agent demos from agents that work.
**Short and practical.** About 2.5 hours of video in 62 focused lectures, plus runnable code for every lecture, so you spend your time building, not watching.
**Running example.** You build agents for the finance team of Kestrel Components, a fictional manufacturer: invoices, vendors, GST, policies and payments. It is realistic enough to show every hard problem and simple enough to follow.
**
Who this course is for
⭐ Python developers who want to build real applications with large language models
⭐ Data scientists and analysts moving into AI engineering
⭐ Engineers who use LangChain, LangGraph or an agent SDK and want to understand what happens underneath
⭐ Professionals in finance, operations and IT who want to automate document and data work with agents
⭐ Anyone preparing for AI engineering interviews
此处内容需要权限查看
会员免费查看
