RAG & Agentic AI Course
From zero to expert. Every topic in three layers: concept → from scratch in Python → real framework. With exercises you run in the browser.
The method: three layers
Concept / design
Why, when, and what it replaces.
From scratch (pure Python)
You implement the mechanism by hand. Stdlib-only, deterministic, runnable here.
Real framework
How it is done in production tools (LangChain, LlamaIndex, CrewAI…).
Modules · 12
Setup & refresher
Environment, offline (mock) mode, Python refresher, and a RAGorbit tour.
LLM & RAG fundamentals
What an LLM is, prompting, the RAG pattern, and embeddings from scratch.
Data ingestion
Loaders, chunking, and metadata: from raw documents to chunks.
Embeddings & vector stores
Real embeddings, ChromaDB, FAISS, and sentence-transformers.
Advanced retrieval
Hybrid search, rerankers, query transformation, and GraphRAG.
Generation, logic & eval
Structured output, citations, and evaluation with RAGAS and metrics.
Agents I — fundamentals
Tool calling, the ReAct loop, memory, Reflexion, and LangGraph from scratch.
Agents II — multi-agent
Multi-agent with LangGraph, CrewAI, AutoGen/AG2, and BeeAI.
Model Context Protocol
MCP server and client with FastMCP, and its security.
Production & security
Guardrails, HITL, observability, deployment, AI security, and UIs.
Multimodal — voice & vision
STT/Whisper, vision, image/audio generation, and multimodal embeddings.
Architecture & capstone
Rebuild templates, a design challenge, and the integrated exam.
RAGorbit
Every topic in the course is anchored to a RAGorbit block and to one of its 10 industry templates: you draw the flow on a canvas and get a deployable Python project, with mocks and tests. By the end of the course you can rebuild those templates from scratch — and design your own.
Free and open source, like the course. One file, zero dependencies.
Knowledge base (vendor-neutral)
Glossary
~138 terms across the RAG/agents ecosystem.
Technologies compared
Comparison tables + an honest critique of the Lang* stack.
RAG without LangChain
The same RAG with LlamaIndex, Haystack, and the native SDK.
Agents without LangChain
The same agent with a native loop, CrewAI, AutoGen/AG2, and Pydantic-AI.
Landscape: databases
The storage market for AI systems.
Landscape: processes
Orchestration, serving, pipelines, and deployment.
Landscape: RAG strategies
~32 advanced RAG techniques and architectures.
Node catalog
Reference for the RAGorbit flow language nodes.
Mapped templates
The 10 industry templates explained and mapped.
IBM/Coursera coverage
Mapping the IBM Coursera syllabus onto the course.