Catalogue · AI & Agents · intermédiaire

Understanding modern RAG

9 modules · 26 lessons ~4h of readingUpdated Apr. 2026

Everyone talks about RAG, very few people know what actually happens between an uploaded document and a displayed answer. This course opens the black box, step by step, in plain language. You understand why a language model generates instead of searching, how a machine represents meaning, why chunking decides everything, what a vector database is for, and what a similarity score really measures. You then discover the techniques that separate a prototype from a reliable system: hybrid search to also retrieve exact codes and references, query transformation, reranking, context compression, strict prompting, permissions applied before retrieval, hallucination detection. Then the advanced architectures (Corrective, Adaptive and Self-RAG, GraphRAG, Agentic RAG), measurable evaluation, cost and latency control, and the full production architecture. All illustrated by an end-to-end example project: the document assistant of Camille, who runs a forty-person Swiss company. No code to write: technical details live in optional collapsible boxes you open only out of curiosity.

~~CHF 160~~CHF 96-40%one-time payment · lifetime access

By the end

What you'll walk away knowing how to do

A real, working result in your hands, not a throwaway demo, something you keep.

The full method to do it again on your own, on your own ideas.

A clear Bronze → Silver → Gold progression: a tangible result at every tier.

Enough to never get stuck: every step explained, nothing assumed.

Is this for you?

Who it's for

Intermediate level

Requirements: a computer, a connection, and the will to move forward. Every step is explained, nothing is assumed.

You want a real result in your hands, not just theory.
You're starting from zero or close to it: no prior technical skill required.
You want to move at your own pace, no live classes, no fixed schedule.

Why this format over the others

vs a video course

Dense, searchable, copy-paste-ready text. Straight to the point, no scrubbing a 40-min video for two commands.

vs free docs

A clear, ordered path from first step to result. No traps, no blind spots, no hours lost piecing it together.

vs a bootcamp

One-time payment, lifetime access, updates included. Move at your own pace, no fixed schedule, no extra fees.

Detailed programme

Introduction

Why RAG exists · what a model really knows · how a machine stores meaning

An LLM doesn't search, it generatesIf there is one single idea to take away from this whole course, it is this one: a language model does not search for information, it produces text. That sentence sounds harmless. Yet it explains almost every limitation you have observed while using a conversational AI.7 minPreview

How a machine represents meaningWe now know why RAG exists. One unsettling question remains: how can a computer "understand" a sentence? The answer is surprising. It does not understand it. It transforms it.7 minPreview

Module 1 — Preparing the knowledge

Document chunking · embeddings · the same-model rule

Module 2 — Storing and retrieving

The vector database · semantic search · metadata, filters and permissions

Module 3 — Improving retrieval

Hybrid search and BM25 · query transformation · reranking · context compression

Module 4 — Generating without inventing

The RAG prompt · citations and contradictions · the four families of hallucinations

Module 5 — Advanced architectures

Corrective, Adaptive and Self-RAG · GraphRAG · Agentic RAG and its guardrails

Module 6 — Measuring and industrialising

Measurable evaluation · latency and costs · the full production architecture

Module 7 — Getting practical

The ten-step practical project · the ten most frequent mistakes

Module 8 — Recap and exam

The six-step recap · the final architecture · conclusion · exam and answer key