<h2>What you leave with</h2>
<p>You leave with a running system and written proof, not a notebook of definitions. That is what you show a teammate or a hiring manager: not a ChatGPT screenshot, but a stack you started, pages you listed, and a scorecard you filled.</p>
<ul class="cd-grid">
  <li><strong>Compose stack.</strong> API, Postgres with pgvector, idle scraper — started by you.</li>
  <li><strong>Two tables.</strong> Pages and chunks, including a 3,072-dimension embedding column, walked on the real files of the clone.</li>
  <li><strong>Your crawl.</strong> Paths you listed on a site you chose, skipped where <code>robots.txt</code> disallows them. Not a sitemap reader.</li>
  <li><strong>Ask on your pages.</strong> Pages and chunks you can list, and one ask that answers from those chunks.</li>
  <li><strong>Sample answers.</strong> The files in <code>responses/</code> from the sample site. The mentorship-pricing file lists the three real plans, from $89.99 for one session — the opposite of the invented price. Your crawl does not write these files, and they are not a second knowledge base.</li>
  <li><strong>Scorecard.</strong> One question you write in chapter 1: was the page stored, is the expected fact in a chunk, did the answer stay inside your stored pages, and which stage failed if a check did not pass.</li>
  <li><strong>Next-layer map.</strong> A chat UI, Redis memory, better chunks, a two-check eval, a richer answer with citations, and nginx — to build next, not in this repo.</li>
</ul>

<h2>How the course is built</h2>
<p>Six chapters. Forty-eight short lessons, six to eight minutes each, about five to six hours in all. <strong>One system, one repo:</strong> <a href="https://github.com/mmmohajer/rag_tutorial" rel="noopener noreferrer">github.com/mmmohajer/rag_tutorial</a>. It ships finished. You clone it in chapter 3 and run it; you do not type the backend from empty.</p>
<ul class="cd-grid">
  <li><strong>Chapters 1–2.</strong> Fit, then the eight RAG stages as one picture. Nothing installed, nothing built.</li>
  <li><strong>Chapter 3.</strong> Git, Docker, and Compose on Mac, Windows, or Linux, then <code>docker compose up -d --build</code> and a green health check.</li>
  <li><strong>Chapter 4.</strong> <code>api/</code> in the order it grew — FastAPI, SQLAlchemy, Alembic, pgvector — on the real lines, with a check on the running stack.</li>
  <li><strong>Chapter 5.</strong> Scrapy on pages you list, honouring <code>robots.txt</code>. Crawl, store, ask.</li>
  <li><strong>Chapter 6.</strong> Grade your evidence, then map what a real visitor still needs. A map, not a second lab: no chat UI, Redis, or nginx is added to the clone.</li>
</ul>
<p>Watch the lesson, then do the file. Each chapter opens what the last one left running. The chat model is <code>gpt-6-astra</code>; the embedding model is <code>text-embedding-3-large</code>. No LangChain lab.</p>

<h2>What you need</h2>
<ul class="cd-grid">
  <li>You know what a database is and what an API is. The course does not teach those words from zero.</li>
  <li>Git, Docker, and Compose — installed in chapter 3. Colima works if it is already running, but the course does not debug it.</li>
  <li>An OpenAI account with billing enabled. <code>GET /</code> does not call the model. From chapter 5, every stored chunk is one embedding call and every ask is one embedding plus one chat call, so keep the path list short and set a spend limit.</li>
  <li>Know that the shipped system message is written for the sample site. It calls itself that site’s assistant and refuses code-style questions, even when the fact is on your pages. Chapter 5 shows where it lives.</li>
  <li>The companion repo above, cloned in chapter 3.</li>
</ul>

<h2>Why this, and why now</h2>
<p>Popular 2026 RAG courses go wide, through a framework or a hosted store:</p>
<ul class="cd-grid">
  <li><em>LangChain- AI Engineering with LangChain &amp; LangGraph</em> on Udemy (more than 52,000 ratings, updated October 2026) teaches RAG on the way to LangGraph agents and MCP.</li>
  <li><em>Hands-On RAG with LangChain</em> (updated September 2026) uses Postgres with pgvector, then adds indexes, reranking, and RAGAS through LangChain.</li>
  <li>DeepLearning.AI’s <em>Retrieval Augmented Generation</em> (updated 23 July 2026) covers BM25, hybrid retrieval, Weaviate, and monitoring.</li>
</ul>
<p>This one goes underneath: one readable crawl-to-answer path in FastAPI, Scrapy, Postgres, and pgvector, run with Docker Compose on your machine, on pages you choose, with no orchestration framework between you and the code.</p>
<p>That foundation is what retrieval roles still assume. Siemens Energy’s AI Retrieval &amp; Agent Platform Engineer posting (5 October 2026) asks for FastAPI, Docker, chunking, embeddings, metadata, reranking, hybrid retrieval, and quality measurement. This course teaches the first five on real code and maps the rest in chapter 6. It does not make you job-ready on its own, and it does not teach reranking, hybrid search, or a formal evaluation harness.</p>
<p>Take it if you want to run and understand every file in a small RAG system before you add a framework. Skip it if you want a LangChain certificate or an advanced retrieval-operations course.</p>

<h2>Run it. Then decide what to add.</h2>
<p>You saw the six chapters. Before you pay, watch the four free lessons — who this is for, what you will learn, the eight stages as one picture, and the capstone scorecard — and read the <a href="https://github.com/mmmohajer/rag_tutorial/blob/master/responses/01-mentorship-pricing.md" rel="noopener noreferrer">saved sample answer</a> on GitHub. If crawl, store, answer, plus a written plan for what a real visitor still needs, is the stack you want, enrol.</p>
<p>Whether you are building a site chatbot or starting toward an AI-engineer role, the outcome is the same: answers drawn from pages you chose, that you can open, inspect, and grade.</p>

<h2 class="cd-hook">The chatbot quoted a price. The live page never had it.</h2>
<p class="cd-lead">Picture this. A visitor books mentorship on a number the chat window invented. The invoice is different. You open the pricing page. That number was never there. Nobody typed a wrong price into the site. Someone wired a general model to a box and shipped it. The model is fluent. It is not informed.</p>
<p>That gap is why this course exists. ChatGPT can answer a lot of public questions. It does not know your business, your docs, or your site. You need an assistant that answers from pages you chose, that you can open and inspect, and that you can prove before a real customer uses it. When you finish, you can open the stored page, see the chunk, and tell a teammate which stage failed. That is the difference between a fluent demo and a system you can debug.</p>

<div class="cd-split">
  <div class="cd-card">
    <h2>Who this is for</h2>
    <p>People who want to become AI engineers and to <strong>build</strong> assistants, not only use them. You already know what a database is and what an API is. You want RAG you can read line by line, not a framework call you cannot explain.</p>
    <ul class="cd-grid">
      <li>You need a bot that answers from support pages, a company site, or docs you trust — and you want to open the stored pages and chunks those replies are drawn from.</li>
      <li>You have seen a LangChain demo and still cannot say what happens between the crawl and the answer.</li>
      <li>You want a foundation you can later grow into a support chatbot or an agent, without pretending this course makes you job-ready on its own.</li>
    </ul>
  </div>
  <div class="cd-card cd-card-muted">
    <h2>Who this is not for</h2>
    <p>If you want ChatGPT productivity tips, a LangChain or LlamaIndex tour, a no-code “chat with PDFs” click-path, or a course that assumes no programming, this is the wrong course. A little Python helps. A little Docker helps. Someone who has never used a terminal will struggle.</p>
  </div>
</div>

<div class="cd-quote">
  <p>By the end, you will be able to say this sentence, and show the running stack that backs it:</p>
  <p class="cd-quote-line">I can crawl pages I chose, store them, and answer from those pages — and I can tell you what to add before a real customer uses it.</p>
</div>

For developers who know APIs and databases: crawl pages you chose, store them in Postgres with pgvector, and answer from them. FastAPI, Scrapy, Docker.

RAG Course: Crawl, Store, and Answer from Your Site

Decide whether this course is for you, confirm the prerequisites, and write a one-sentence acceptance test for the system you will build: a named page on a site you choose, one question that page can answer, and the evidence you expect to inspect after the crawl. Nothing is installed here. Keep that acceptance card; chapter 5 completes it.

Start Here: Fit, Requirements, and Outcome

Who this course is for

What you need before you start

What you will learn

Your six-chapter build map

Use one question from your acceptance card to trace all eight RAG stages. For each stage, record its input, output, choice, and likely failure in a one-page pipeline sheet. You do not install or build here; the completed sheet is the chapter win and the diagnostic map you will reuse in chapters 4–6.

RAG in Eight Stages

The eight stages as one picture

Sources become text

Chunk strategy, size, and overlap

Embeddings, and one model for chunks and questions

Vector storage, indexes, and metadata

Cosine similarity, top-k, and the threshold

The system message, the context, and the model call

The answer the person sees, then your pipeline sheet

The first artefact on your machine. Git, Docker on your OS, clone `rag_tutorial`, the three env files, `docker compose up -d --build`, then a high-level folder map. You do not crawl in this chapter.

Run the Compose Stack

Install Git

Docker on a Mac

Docker on Windows

Docker on Linux

Clone the repository

Create the three env files

Start the stack and prove it is up

The folders, after it is running

You cloned the finished repo and started it in chapter 3. This chapter walks `api/` and Postgres in the order the folder grew, on the real lines, so you can explain and change every file. Four checkpoints on your running stack: Compose connects `api` and `db`; the container waits, migrates, and serves; the pgvector tables exist; the six `/rag` routes answer. Every lesson ends with one command or request you run to see that layer working. The scraper stays idle.

The FastAPI and pgvector Backend, File by File

API and Postgres in Compose

The API Dockerfile, line by line

requirements.txt

How the container starts

Turn env values into settings

The tables, in Python

Where the Alembic folder came from

Write, inspect, and apply the migrations

Wire the four HTTP layers

Why each `/rag` route exists

The OpenAI client

The scraper is how rows get in. Point it at a site you choose. Honour `robots.txt`. Crawl. List pages and chunks. Ask one question on your pages. Then read `responses/` as proof the same path already worked on the sample site.

Crawl Your Pages and Ask

The scraper service and image

Libraries, and the Scrapy project file

Let `robots.txt` refuse a page

Choose the pages on your site

Make the spider call the API

Turn one page into numbered chunks

Run the crawl

Prove the knowledge base, then ask

Saved answers from the sample site

Capstone first, roadmap second. Open on pride, then grade the evidence from chapter 5 — listed pages, listed chunks, your answer, and the saved sample answers — on a pass/fail scorecard. Then map the next layers a real visitor needs and close on pride. Nothing here is built in this repo: no chat UI, Redis, nginx, hybrid search, or eval harness.

Grade the Pipeline, Then Map Production

Capstone: grade your crawl-store-ask pipeline

A chat a person can use

Memory, with Redis

Chunks can get better

How you know which stage failed

The answer the person sees

Nginx is the front door

Your next production backlog

RAG You Can Run: Crawl Your Site, Store the Pages, Answer from Them

<div class="course-slide cs-center">
  <p class="cs-kicker">Monday morning</p>
  <h2 class="cs-title cs-at-2-7">A price the chat window quoted.</h2>
  <p class="cs-pill cs-at-7-4">That number was never there.</p>
</div>

<div class="course-slide">
  <p class="cs-kicker">What actually happened</p>
  <h2 class="cs-title">Nobody typed a wrong price.</h2>
  <div class="cs-row cs-at-2-6">
    <span class="cs-chip">general model</span>
    <span class="cs-chip">+ chat box</span>
    <span class="cs-chip">shipped</span>
  </div>
  <div class="cs-grid cs-cols-2">
    <div class="cs-card cs-at-6">
      <p class="cs-card-label">Fluent</p>
      <p class="cs-card-text">Sounds like your site.</p>
    </div>
    <div class="cs-card cs-card-danger cs-at-7-3">
      <p class="cs-card-label">Informed</p>
      <p class="cs-card-text">Never read your <span class="cs-bad">pricing page</span>.</p>
    </div>
  </div>
</div>

<div class="course-slide cs-center">
  <p class="cs-kicker">What this costs you</p>
  <h2 class="cs-title cs-at-4-6">A ChatGPT screenshot. Nothing you can debug.</h2>
  <p class="cs-pill cs-at-7-4">Looks finished. Until the first real question.</p>
</div>

<div class="course-slide">
  <p class="cs-kicker">RAG · retrieval-augmented generation</p>
  <h2 class="cs-title cs-at-3">Closes that gap.</h2>
  <p class="cs-text cs-at-4-2"><span class="cs-strike">A tour of wrappers</span></p>
  <div class="cs-card cs-card-accent cs-card-glow cs-at-5-7">
    <p class="cs-card-title">One small system. Your machine.</p>
    <p class="cs-card-text cs-at-9-7">Answers from <span class="cs-hl">pages you chose</span> and can open.</p>
  </div>
</div>

<div class="course-slide">
  <p class="cs-kicker">Read it before you pay</p>
  <h2 class="cs-title cs-at-3">A saved answer in the companion repo.</h2>
  <p class="cs-note cs-mono cs-at-3">rag_tutorial/responses/01-mentorship-pricing.md</p>
  <p class="cs-quote cs-at-6-6">What is your mentorship pricing?</p>
  <div class="cs-grid cs-cols-3 cs-at-9-1">
    <div class="cs-card cs-card-accent">
      <p class="cs-stat-value">$89.99</p>
      <p class="cs-stat-label">1 session</p>
    </div>
    <div class="cs-card">
      <p class="cs-stat-value">$349.95</p>
      <p class="cs-stat-label">5 hours</p>
    </div>
    <div class="cs-card">
      <p class="cs-stat-value">$549.90</p>
      <p class="cs-stat-label">10 hours</p>
    </div>
  </div>
</div>

<div class="course-slide cs-center">
  <p class="cs-kicker">The saved answer</p>
  <h2 class="cs-title cs-at-0-8">One session. <span class="cs-hl">$89.99</span></h2>
  <p class="cs-pill cs-at-3">Unused sessions never expire.</p>
  <p class="cs-pill cs-at-5-5">This time, the price came from the page.</p>
</div>

<div class="course-slide">
  <p class="cs-kicker">Then you test it</p>
  <h2 class="cs-title">A short scorecard on a site you pick.</h2>
  <ul class="cs-checks">
    <li class="cs-at-4-4">The page was stored</li>
    <li class="cs-at-5-3">The fact is in a chunk</li>
    <li class="cs-at-6-6">The answer stayed inside your pages</li>
  </ul>
  <p class="cs-pill cs-at-9">When a check fails, you name the stage.</p>
</div>

<div class="course-slide cs-center">
  <p class="cs-kicker">Defend it in review</p>
  <h2 class="cs-title cs-at-1-1">A stack you can defend.</h2>
  <p class="cs-pill cs-at-2-7">Stay: future AI engineer, or a bot on your own site.</p>
</div>

<div class="course-slide">
  <p class="cs-kicker">On real code</p>
  <h2 class="cs-title">FastAPI. Docker. Chunking. Embeddings.</h2>
  <p class="cs-text cs-at-3-5">The foundation those roles still assume.</p>
  <div class="cs-grid cs-cols-2 cs-at-6-7">
    <div class="cs-card cs-card-accent">
      <p class="cs-card-label">Stay</p>
      <p class="cs-card-text">You want to build the assistant.</p>
    </div>
    <div class="cs-card cs-card-danger">
      <p class="cs-card-label">Skip</p>
      <p class="cs-card-text">ChatGPT tips. LangChain tour. No-code. New to a terminal.</p>
    </div>
  </div>
</div>

<div class="course-slide cs-center">
  <p class="cs-kicker">Not job ready by itself</p>
  <h2 class="cs-title cs-at-2-5">Your next project, <span class="cs-hl">better</span>.</h2>
  <p class="cs-pill cs-at-5">Lifetime access.</p>
  <p class="cs-pill cs-at-6-7">Four free lessons.</p>
  <p class="cs-pill cs-at-12-5">Then enrol.</p>
</div>