Modern AI Engineering

Lesson 1.1 · 23 min · video lesson

Six Concepts Every AI Engineer Must Know

Six words show up in almost every AI job post and product demo: LLM, RAG, MCP, Agent, Fine-tuning and Quantization. What does each one actually do, and how do they click together into one working product?

In short: An LLM is a model that predicts text; RAG feeds it fresh facts at question time; MCP is a standard plug that connects it to tools and data; an Agent lets it act in a loop; Fine-tuning changes its habits by training; Quantization shrinks it so it runs cheaply. We tour all six through one running example: a support assistant for an online shop.

What you will learn

  • What an LLM is
  • What RAG is
  • What MCP is
  • What an Agent is
  • What Fine-tuning is
  • What Quantization is
  • How the six fit together in one real product

In this lesson

  1. Why start with six words?
  2. Word 1: LLM, the engine that predicts text
  3. Word 2: RAG, giving the model the right facts
  4. Word 3: MCP, a standard plug for tools and data
  5. Word 4: Agent, an LLM that can act in a loop
  6. Word 5: Fine-tuning, changing the model's habits
  7. Word 6: Quantization, making the model smaller and cheaper
  8. Try it: toy RAG and toy quantization in Python
  9. How the six fit together in one product
  10. Common mistakes and how to spot them
  11. Practice: try it yourself

Key terms in this lesson

LLM
A large neural network trained on text to predict the next token, used to understand and generate language.
Token
A small unit of text (a word or part of a word) that a language model reads and writes.
RAG
Retrieval-Augmented Generation: retrieving relevant documents and adding them to the prompt before the model answers.
MCP
Model Context Protocol: an open standard for connecting AI applications to tools, data sources and prompt templates.
Agent
A system in which an LLM repeatedly decides on actions, uses tools, and observes results until a goal is met.
Fine-tuning
Further training a pre-trained model on focused data so its weights, and therefore its behaviour, change.
Quantization
Storing model weights with fewer bits (for example 8 or 4) to reduce memory and speed up inference.

This lesson is free. Sign in or create a free account to open it with its video and quiz.

2.1 Machine Learning from First Principles →