Generative AI Course for Beginners: What to Learn, in Order
By Modern AI Engineering · · 8 min read
A good generative AI course for beginners teaches six things in order: what generative AI is, how a large language model produces text, how to prompt it, how to give it your own data with RAG, how to let it act through agents, and how to test and run what you built. Each step makes the next one easier.
This guide explains that order, says what you should be able to do after each step, and helps you tell a course for people who use AI tools from a course for people who build with them.
What is generative AI, and what does a course on it teach?
Generative AI is the part of AI that creates new content: text, images, audio, video or code. A chat assistant that writes an email is generative AI. So is a tool that draws a picture from a description.
Courses with this name come in two kinds. The first kind teaches you to use the tools well: how to write prompts, where the tools help at work, where they go wrong. The second kind teaches you to build products with the models. That means calling a model from code, connecting it to data, and checking the quality of its answers.
Both are useful, but they are different courses. Before you start one, decide which you need. The rest of this article is about the second kind, the one for people who want to build.
What do you need before you learn generative AI?
For a building course you need basic Python: variables, functions, loops, lists, dictionaries and calling an API. You also need school-level math. You should be comfortable with a graph, an average and a simple probability.
You do not need a degree in machine learning. But a short pass through the basics helps a great deal. If you know what training means, what a loss is and what overfitting looks like, the rest of the course stops feeling like magic.
If you have never programmed, learn Python first. A few weeks of practice is enough to follow most lessons.
What to learn in generative AI, in order
This is the order that works for most beginners. Do not skip ahead to agents because they sound exciting. Agents are built from every step before them.
- The vocabulary. Learn what LLM, token, prompt, context window, RAG, agent and fine-tuning mean, at the level of one sentence each.
- Foundations. Learn how a model learns from data and what a neural network is. Keep it short and focused.
- How an LLM works. Tokens, embeddings, attention and the Transformer. Then how text is produced one token at a time, and what temperature does.
- Prompting. Clear instructions, examples, output formats, breaking a task into steps.
- RAG. Chunking documents, embeddings, vector search, reranking, and answering with sources.
- Agents. Function calling, the agent loop, memory, and standards such as MCP.
- Fine-tuning. What it changes, how LoRA makes it affordable, and when you do not need it.
- Evaluation and safety. Test sets, model-graded checks, guardrails and prompt injection.
- Serving. Response time, cost, caching, quantization and system design.
Do you need machine learning before generative AI?
You need a little. You can call a model through an API on your first day with no theory at all, and that is a fine way to stay motivated. The trouble comes later. When the answers are wrong, you have to work out whether the fault lies in the prompt, the retrieved text, the sampling settings or the model.
That reasoning rests on a few ideas from machine learning and deep learning. Training, loss, overfitting, gradient descent and the shape of a neural network are the main ones. A beginner can cover them in a few focused weeks. After that, tokens, embeddings and attention are much easier to follow.
Other kinds of generative model, such as diffusion models for images, are worth a look as well. You will mostly build with LLMs, but it helps to know that text generation is not the only method.
What should a generative AI syllabus include?
Use this table to check a syllabus. A course that stops after the third row teaches you to make a demo. The later rows are what turn a demo into a product.
| Topic | What you should be able to do afterwards |
|---|---|
| How LLMs work | Explain tokens, embeddings, attention and next-token prediction in your own words. |
| Generation settings | Choose temperature and sampling settings for a task and explain the effect. |
| Prompting | Write a prompt with instructions, examples and a fixed output format, and test it. |
| RAG | Build a question-answering tool over documents that shows its sources. |
| Agents | Write an agent loop with a few tools, a step limit and a log. |
| Fine-tuning | Say when fine-tuning is the right choice and describe how LoRA works. |
| Evaluation | Build a test set and measure whether a change made answers better. |
| Safety | Add guardrails and limit the damage of prompt injection. |
| Serving | Reason about cost and response time, and name ways to reduce both. |
What should you build while you learn?
Build one small thing after each step. Small projects that you understand fully teach more than a large one copied from a tutorial.
- A script that sends a prompt to a model and prints the answer. Then change the temperature and compare the results.
- A summariser that always returns the same structure, such as a title, three points and one open question.
- A question-answering tool over a folder of your own documents, with the source shown for each answer.
- A list of twenty test questions for that tool, with the answers you expect. Run it after every change.
- An agent with two tools, for example search and a calculator, that stops after a fixed number of steps.
How long does it take to learn generative AI?
It depends on your starting point and the hours you can give. Someone who already programs and studies for an hour or two a day can cover the core ideas in a few months. Someone new to programming needs longer, because programming takes practice of its own.
Treat promises of mastery in a few days with care. You can learn to call an API in an afternoon. Learning to build something reliable takes longer, and most of that time goes into testing and fixing.
A better measure than hours is this: can you explain why your system gave a wrong answer, and do you know what to change?
How the AI Engineering Bootcamp teaches generative AI
The AI Engineering Bootcamp on this site has 149 video lessons in 19 modules. It follows the order above. Modules 2 and 3 cover machine learning and neural networks. Module 4 opens with a lesson on what generative AI is, then goes inside the Transformer. Later modules cover generation, prompting, RAG, agents, inference, evaluation and safety.
There are three tracks. The Generative AI Engineering track covers LLMs and the systems around them. The ML and Deep Learning track covers the foundations. The Complete AI Engineer track includes both, and adds the certificate of completion.
The first lesson is free with a free account. It explains six words you will meet in every later module: LLM, RAG, MCP, agent, fine-tuning and quantization.
Frequently asked questions
Can a beginner learn generative AI?
Yes. With basic Python and school-level math you can start. Learn the topics in order, from how a model works to prompting, RAG and agents, and build something small at each step.
Do I need to know coding to learn generative AI?
To use AI tools, no. To build products with them, yes. Python is the usual language, and functions, loops, lists and dictionaries are enough to begin.
What is the difference between a generative AI course and a machine learning course?
A machine learning course teaches you to train models on data. A generative AI course teaches you to build with models that already exist, mostly LLMs. The second relies on ideas from the first.
Should I learn RAG or agents first?
RAG first. Retrieval is simpler to build and to test, and most agents use retrieval as one of their tools.