Prompt Engineering
Anyone who already uses AI tools daily and wants reliable results instead of occasional good ones
Prompt Engineering is an online course on getting reliable results from language models: what a prompt actually is, how context drives output quality, and the patterns that survive real use. It ends with a build project and a certificate.
The craft of getting dependable work out of a language model, and the discipline of proving it. You learn what the model is doing when a prompt succeeds, how to diagnose one that is nearly right, how to write output contracts that hold, and how to measure quality with a test set instead of an impression. You finish with a measured prompt, a versioned library you actually use, and a documented rescue of a prompt that was failing.
2 hours reading + 11.5 hours of projects
What you'll be able to do
- Explain why the same prompt produces different answers, and design around it
- Construct a prompt from its five working parts instead of rewriting until something works
- Decide when an example will outperform any amount of instruction
- Write output contracts, including a refusal path the model can actually take
- Diagnose a prompt that is nearly right, by isolating which part is failing
- Measure a prompt with a real test set and report its variance honestly
- Build a versioned prompt library that survives being handed to somebody else
- Recognise the point where prompting stops and retrieval, tools or code must take over
Modules in this course
Why the same words produce different answers, what the model is really doing with your instruction, and why this makes prompting a craft with rules rather than a set of magic phrases.
9 min read
Role, task, context, constraints and output contract: what each part is for, which one people always omit, and how to assemble them instead of rewriting until something works.
10 min read
Why adding more material can make an answer worse, how the model weighs what you supply, and the discipline of giving it exactly what the task requires.
9 min read
Why the model fills silence with the statistically ordinary, how to find the assumptions hiding in your own request, and what to specify before the first run.
9 min read
Why one worked example can outperform a paragraph of instruction, how many to supply, and the mistake that teaches the model exactly the wrong pattern.
9 min read
Why one prompt doing five jobs fails, how to split a task into checkable stages, and when asking for reasoning helps rather than merely producing more text.
10 min read
How to specify a response shape that holds across runs, why every contract needs a defined failure output, and what to do when the format still drifts.
10 min read
A repeatable method for finding which part of a prompt is failing, instead of rewriting the whole thing and hoping.
10 min read
How to build a test set for a prompt, why every case must run several times, and how to report quality as a number you can defend.
10 min read
Turning a prompt that worked once into an asset: parameterising it, storing it where it can be found, versioning changes, and knowing when to retire one.
9 min read
The five task shapes that cover most professional use, what each one gets wrong by default, and the specific correction for each.
10 min read
The four boundaries no prompt can cross, what each one requires instead, and how to recognise you have hit one before spending a week iterating.
9 min read
Why any text your prompt reads can carry instructions, the defences that actually work, and the confidentiality rules that apply the moment you paste somebody else's material.
9 min read
Cost and model choice, keeping prompts alive as models change, teaching the method to a team, and the habits that separate sustained skill from a good week.
9 min read
A prompt you actually use, rebuilt from its five parts, with a 20-case test set run three times each and an honest written result including its failures.
210 min to build
Five parameterised, documented and measured prompts in one location, versioned, with the test sets stored beside them.
240 min to build
Take something genuinely broken, diagnose it by isolating one change at a time, and document the rescue with before and after numbers, including what could not be fixed.
240 min to build
Questions before you buy
- How long does this course take?
- About 2 hours of reading, plus roughly 11.5 hours across the 3 build projects. Most of that time is the build projects, at your own pace.
- Do I need any experience first?
- No formal prerequisites — every course is written to be readable from the first lesson, and that lesson is free to read before you buy, so you can check it's pitched right for you.
- How is the certificate issued?
- Automatically, the moment you complete every module and project. No review queue to wait on.
- What's your refund policy?
- A 14-day money-back guarantee, no questions asked — email support@courseparse.com within 14 days of purchase. Full policy on the Refunds page.