Free AI & Prompt Engineering Course with Certificate
A free, self-paced AI and prompt engineering course with flash cards, checkpoint quizzes and a verifiable completion certificate. Learn LLM fundamentals, prompt patterns, RAG, evaluation and safe AI use at work.
A large language model (LLM) is a neural network trained to predict the next token in a sequence. It has no database of facts it looks things up in; it has statistical patterns learned from enormous amounts of text. Everything it produces — an essay, code, an apology — is the same operation repeated: pick a likely next token, then do it again.
An LLM predicts text. It does not retrieve truth.
Ask any chatbot the same question twice in new chats and compare the wording. Same meaning, different tokens — that's prediction, not lookup.
Every expert you admire learned it exactly this way — one scene at a time.
What you will be able to do
- Explain how large language models generate text, and where they fail
- Write structured prompts that produce consistent, usable output every time
- Apply role, context, constraint, format and example patterns deliberately
- Design multi-step and chain-of-thought style workflows for complex tasks
- Ground answers in your own documents with retrieval (RAG) thinking
- Evaluate, test and version prompts like a professional, not by vibes
- Spot hallucinations, bias, prompt injection and privacy risk before they cost you
- Ship a reusable prompt library your whole team can run
Course structure
Foundation — How AI language models actually work
The mental model behind every chatbot: tokens, context, probability, and why a confident answer can still be wrong.
Advanced — Patterns, grounding and reliable workflows
Move from single prompts to systems: decomposition, reasoning patterns, retrieval, structured output and prompt chains that hold up under pressure.
Mastery — Evaluation, governance and shipping AI at work
The professional layer: measuring quality, controlling cost, managing risk, and rolling AI out to a team without breaking trust.
Full syllabus — every topic covered
30 topic cards across three levels. Each card teaches one idea, ends with a takeaway, and is followed by a checkpoint quiz every five cards.
Level 1: Foundation — How AI language models actually work
The mental model behind every chatbot: tokens, context, probability, and why a confident answer can still be wrong.
- 1. What a large language model really is — An LLM predicts text. It does not retrieve truth.
- 2. Tokens: the unit AI charges and thinks in — Fewer, better tokens means cheaper, faster and usually more accurate output.
- 3. The context window is the model's whole world — If it isn't in the context window, the model cannot use it.
- 4. Training data has a cutoff — and consequences — For anything recent or proprietary, supply the facts yourself or use a tool that fetches them.
- 5. Hallucination: why confident answers can be false — Treat every factual claim, number, quote and link as unverified until you check it.
- 6. Temperature and why answers vary — Low temperature for accuracy and consistency, higher for ideas.
- 7. The anatomy of a strong prompt — Role, task, context, constraints, format. Name all five and guesswork disappears.
- 8. Show, don't just tell: few-shot prompting — Two good examples beat two paragraphs of description.
- 9. Specify the output format on purpose — The shape of the answer is your job to define, not the model's to guess.
- 10. Iteration beats the perfect first prompt — Prompting is a conversation loop, not a magic incantation.
Level 2: Advanced — Patterns, grounding and reliable workflows
Move from single prompts to systems: decomposition, reasoning patterns, retrieval, structured output and prompt chains that hold up under pressure.
- 1. Task decomposition beats one giant prompt — Chain small verified steps instead of gambling on one big answer.
- 2. Reasoning prompts and reasoning models — Give the model room to reason, then check the conclusion independently.
- 3. The system prompt sets the standing rules — Standing rules go in the system prompt; the specific ask goes in the user turn.
- 4. Grounding with retrieval (RAG), in plain English — RAG fixes 'the model doesn't know our stuff' — it doesn't fix bad source documents.
- 5. Fine-tuning vs prompting vs RAG — Prompt → RAG → fine-tune, in that order of escalation.
- 6. Structured output and schema thinking — Define the schema, including the 'unknown' case, before you send the prompt.
- 7. Self-critique and evaluator prompts — Generate, then evaluate against a written checklist. Two passes, two roles.
- 8. Prompt injection and untrusted content — Anything the model reads is untrusted input, never instruction.
- 9. Privacy, confidentiality and what not to paste — Match the sensitivity of the data to the terms of the tool you're using.
- 10. Building a reusable prompt library — A prompt used twice should become a template with a version number.
Level 3: Mastery — Evaluation, governance and shipping AI at work
The professional layer: measuring quality, controlling cost, managing risk, and rolling AI out to a team without breaking trust.
- 1. Build an evaluation set before you optimise — If you can't measure it on fixed cases, you're not engineering — you're guessing.
- 2. Choosing metrics that mean something — Match the metric to the failure you actually fear.
- 3. LLM-as-judge, and its limits — Automated grading scales; human spot-checks keep it honest.
- 4. Cost, latency and model selection — Right-size the model per step instead of using the biggest one everywhere.
- 5. Human-in-the-loop by risk tier — Automate the work, keep the accountability human where it matters.
- 6. Bias, fairness and representational harm — Test for uneven performance across groups before deployment, not after complaints.
- 7. Disclosure, copyright and the rules you work under — Know your organisation's disclosure rule and add meaningful human authorship to anything you need to own.
- 8. Monitoring in production — Ship with logging, a rollback path and a scheduled re-evaluation.
- 9. Rolling AI out to a team — Two workflows done properly beat a company-wide licence nobody uses.
- 10. Staying current without drowning — Fundamentals compound; product news depreciates. Re-test, don't re-learn.
Frequently asked questions
- Is this prompt engineering course really free?
- Yes. Every card, checkpoint quiz, the final test and the completion certificate are free on Moyan AI. You only need a free account to save progress and receive the certificate.
- Do I need coding experience?
- No. The Foundation and Advanced levels need no code. Mastery mentions APIs and evaluation tooling conceptually, but you can complete and pass without writing code.
- How long does the course take?
- About 9 hours of focused study across three levels. Most learners finish in one to two weeks doing a few flash cards a day.
- How does the final test work?
- You get two attempts per seven-day window. Score 70% or higher to pass. If both attempts fail, the test unlocks again the following week so you can revise first.
- Is the certificate verifiable?
- Yes. Each certificate carries a unique ID, the total course duration and a QR code that resolves to a public verification page, and you can share it to LinkedIn and other platforms in one click.