Course Content
Prompt Engineering Mastery
6 sections · 32 lessons
What is prompt engineering?
What you need to know
A prompt is not just the question you type. For an application, it is the whole input the model receives on one call.
The parts you control
| Part | What it does | Example |
|---|---|---|
| System prompt | Sets role, rules and tone for the whole conversation | "You write product descriptions for an Indian home-goods store." |
| Instruction | Says what to do on this call | "Write a 60-word description." |
| Context | Facts the model cannot know | Product specs, brand voice guide |
| Examples | Show the pattern you want | Two approved past descriptions |
| Output format | Shape of the answer | JSON with title and body |
| Settings | How the model generates | Reasoning effort, max tokens, a JSON schema |
Why it is a real skill
A model does exactly what the text implies, not what you meant. A vague prompt leaves dozens of decisions — length, audience, format, what to leave out — to the model, and it makes them differently each time. A precise prompt makes those decisions for it. When the output feeds another system (a database, a web page, a support queue), that consistency is the whole point.
What changed by 2026
Older models needed tricks: "think step by step", ALL-CAPS warnings, prefilled answers. Current reasoning models plan and reason on their own, follow instructions more literally, and APIs now offer features that replace many tricks — structured outputs for JSON, reasoning-effort settings instead of "think harder". So the skill has shifted towards context engineering: choosing what information goes into the window, stating the goal and constraints clearly, and measuring the result.
A real-life example
An e-commerce site selling kitchenware needs 5,000 product descriptions. The first prompt:
Write a product description for this item: Steel pressure cooker 5L.The output is 250 words of generic praise, invents a "lifetime warranty", and uses American spelling. Every run gives a different length. The engineered prompt:
You write product descriptions for HomeKart, an Indian kitchenware store.Readers are home cooks shopping on their phones.Write one description for the product below.- 50 to 70 words, British/Indian English, no exclamation marks.- Use only the facts in <specs>. Do not mention warranty, price or delivery unless they appear in <specs>.- End with one line on who the product suits.<specs>Name: Steel pressure cooker | Capacity: 5 L | Material: 304 stainless steelBase: induction-compatible | Safety: gasket release valve</specs>Now the output is a consistent 60-odd words, mentions induction compatibility, and never invents a warranty. The team saves the prompt in Git as product_desc_v3, runs it on 50 held-out products before each change, and a reviewer checks a random 2% sample weekly.
Follow-up questions to expect
- "Is prompt engineering still needed now that models are so capable?" — Less trickery is needed, but the model still cannot know your audience, business rules or output contract. Specifying those, and testing them, is permanent work.
- "When would you fine-tune instead?" — When you need a behaviour at very high volume that prompting cannot reach reliably, or want a smaller, cheaper model to match a larger one. Try prompting and retrieval first; they are cheaper to change.
- "What is context engineering?" — The broader job of deciding what goes into the context window on every call — instructions, retrieved documents, tool results, memory — and what to leave out so the important parts are not buried.