Pebellslearn by building
RoadmapsMatrixCoursesSandboxSign in

Course details

Prompt Engineering

Learn how large language models turn text into text, and how to steer them on purpose. You will control randomness and length, use zero-shot, few-shot and reasoning techniques, and test prompts for bias, overconfidence and attack — all on one running task: a support inbox for a small bike shop.

Ready to learn6 sections · 24 lessons
Start learning

Course curriculum

Introduction3 lessons

See a language model as a next-token predictor, a prompt as everything the model reads before it answers, and prompt engineering as the habit of designing and testing that input on purpose.

  1. LLMs — Predicting the Next Token12 min
  2. What Is a Prompt?12 min
  3. Prompt Engineering — Design, Test, Revise12 min
How LLMs Behave5 lessons

Tokens and the context window, what model weights store, why models hallucinate, when to prompt versus retrieve versus fine-tune, and how agents open the door to prompt injection.

  1. Tokens and the Context Window12 min
  2. Model Weights — What the Model Knows12 min
  3. Hallucination — Fluent but False14 min
  4. Prompting, RAG or Fine-Tuning?15 min
  5. Agents and Prompt Injection15 min
Controlling the Output4 lessons

The settings that shape what comes out beyond the prompt text: temperature, top-k and top-p sampling, max tokens, stop sequences, and frequency and presence penalties.

  1. Temperature — The Randomness Dial12 min
  2. Top-K and Top-P — Trimming the Choices14 min
  3. Max Tokens and Stop Sequences12 min
  4. Frequency and Presence Penalties12 min
Core Prompting Techniques3 lessons

Zero-shot prompts with sharp instructions, one-shot and few-shot prompts that teach by example, and system, role and contextual prompts that set the frame for everything else.

  1. Zero-Shot Prompting12 min
  2. One-Shot and Few-Shot Prompting14 min
  3. System, Role and Contextual Prompts15 min
Prompting for Reasoning5 lessons

Chain-of-thought and step-back prompting for multi-step problems, self-consistency for more reliable answers, tree of thoughts for exploring options, and ReAct for mixing reasoning with tool use.

  1. Chain-of-Thought Prompting14 min
  2. Step-Back Prompting12 min
  3. Self-Consistency and Ensembling14 min
  4. Tree of Thoughts14 min
  5. ReAct — Reason, Act, Observe15 min
Testing and Improving Prompts4 lessons

Let a model search for better prompts, have models grade outputs, keep confidence honest, remove bias from examples, and red-team the whole system before it meets real customers.

  1. Automatic Prompt Engineering14 min
  2. LLM Self-Evaluation14 min
  3. Calibration and Debiasing15 min
  4. Red-Teaming Your Prompt15 min