AI Engineering Fundamentals Build LLM Apps, RAG and Agents

Build working LLM apps with prompts, structured outputs, RAG, tool calling and AI agents, then test them before users do
What you'll learn

---

Map any AI product onto the layers of an LLM app and explain how an LLM generates text token by token
Call LLM APIs reliably with chat history, streaming, retries and safely stored API keys, or run open models locally with Ollama
Write system prompts, reasoning prompts and versioned templates that return schema-valid JSON from text, images and PDFs
Extract, classify and summarize thousands of documents with validation, confidence routing and cited claims
Build a RAG chatbot over your own PDFs with vector databases, hybrid search, reranking and citations
Evaluate RAG by measuring retrieval and answers separately
Build tool-calling agents and workflows with memory, step limits, spend caps and human approval
Build a research agent with web search and decide when frameworks like LangChain or LlamaIndex are worth it
Test LLM apps with eval sets, a checked LLM judge, agent path evals and evals that run in CI on every change
Trace every call, redact personal data, block prompt injection and cut cost with caching, batching and smaller models
Improve prompts from real users with feedback loops and A/B tests
Explain MCP, multi-agent systems, fine-tuning, voice, image generation and computer-use agents, and plan a portfolio of three projects
Requirements

---

A computer with Python 3 installed and an API key from one LLM provider (a few US dollars of credit is enough)
You can read simple Python: variables, functions, lists and dictionaries. Every line of course code is explained
No machine learning, math or framework experience needed
Description

---

This course contains the use of artificial intelligence.
You have used ChatGPT or Claude. Now you want to understand how they work and build with them. This course takes you from typing prompts into a chat box to working LLM apps that answer questions from your own documents, call tools, run as agents, and pass test sets you wrote yourself. Every idea is explained in plain words first, with a picture and a real example, so you are never left guessing what a term means.
It starts with a map of the whole field. In the first lesson you see every layer of an LLM app on one screen: the model, prompts and context, retrieval, tools and agents, evals and guardrails, and the app itself. Then you climb one layer at a time, and every section ends with something you can check.
This course is for beginners: developers new to AI, students, analysts, product people and career changers who can read simple Python. You do not need machine learning, math or any framework. It is not for engineers who already ship RAG and agents in production, and it is not a course on training models.
What makes it different: most beginner AI courses show a demo that worked once. Here every hard idea gets a drawn diagram, two examples and a counterexample. You learn to test your app with evals, run them in CI, trace every call, catch hallucinations and block prompt injection, the parts that decide whether an AI feature survives real users.
What is inside:Section 1, How LLMs actually work: the AI engineering stack, next-token prediction, temperature, tokens and context windows, your first API call, chat history, streaming with rate limits and retries, running open models locally with Ollama, and choosing a provider while keeping API keys safe. Section 2, Prompts and structured outputs: system prompts, few-shot examples, JSON Schema outputs, fixing broken prompts, reasoning prompts, images and PDFs as input, prompt templates and versioning, extraction and classification at scale, and summarization that keeps the facts. Section 3, RAG: embeddings, chunking, a full retrieve, augment, generate pipeline, RAG debugging, vector databases, hybrid search and reranking, a project where you build a RAG chatbot over your PDFs, and evaluating retrieval and answers separately. Section 4, Tool calling and AI agents: tool calls, the agent loop, prompt chaining versus agents, guardrails, routing, parallel calls and evaluator loops, agent memory, a project where you build a research agent with web search, and when agent frameworks like LangChain and LlamaIndex are worth it. Section 5, Testing, securing and improving LLM apps: hallucinations, eval sets, LLM-as-a-judge, prompt injection, observability, PII and content guardrails, cutting cost and latency, evals in CI, agent evals, user feedback and A/B tests, and the capstone where you ship a document Q&A assistant with an agent. Section 6, Extra: MCP, context engineering, choosing reasoning, multimodal or open models, coding agents, multi-agent systems, fine-tuning versus RAG versus prompting, shipping to production, voice AI apps, image generation, computer-use and browser agents, and a final lesson on building a portfolio of three projects that prove your skills.
Every section starts with a short key terms lecture and ends with a module handout: terms, diagrams, exact code, expected output, and self-check questions with answers. Each section also has a quiz and a 10 to 15 minute assignment.
What this course does not cover: training models from scratch, the math behind transformers, and deep MLOps. Fine-tuning, frameworks and production are covered at the level you need to decide when to use them.
Who this course is for

---

Beginners who want to understand how LLM apps, RAG and AI agents actually work and build one themselves
Developers, students, analysts and product people starting in AI engineering
Not for you if you already ship RAG or agents in production, or if you want model training, fine-tuning or MLOps
Homepage

---