— free · one lesson a day · no signup

Practice prompt engineering,
with a score on your prompt

Not another video course. You write the prompt for a real task, an AI judge scores each criterion from 0 to 100 and shows exactly what was missing.

~3 min per lesson · techniques with real sources · daily streak
24lessons in the library
5competency tracks
11sourced techniques
Today's lesson · #18free · no signup

Product Description Without Exaggeration

technique: Negative constraintsMedium4 scored criteria

A new lesson every day at midnight UTC. Today's is always open.

Start now

How the practice works

1

Get a real task

An actual workplace scenario: fintech, support, data, legal. With one named technique to apply and the scoring criteria visible before you write.

2

Write and get scored

A model generates the answer with your prompt and an AI judge scores each criterion from 0 to 100. Not pass/fail: you see where the prompt almost got there.

3

Refine with direction

Three tries per lesson. Each comes with a suggested tweak and a mini-lesson on the principle. At the end, a commented reference prompt shows how a strong one solves it.

The curriculum

Explore the tracks →

Five competency tracks, from clarity to guardrails. Every node is a lesson with a technique, a judge, and a reference answer.

01
Fundamentals: clarity and format
Precise instructions, output format, constraints, and persona — the base of every reliable prompt.
12344 lessons
Clear, specific instructionsStructured output formatNegative constraints+1
02
Context and examples
Few-shot, delimiters, and persona to guide the model through context.
1234566 lessons
Few-shot (examples)Delimiters / XML tagsRole prompting (persona)
03
Reasoning
Chain-of-thought and decomposition for tasks that require thinking in steps.
1234566 lessons
Chain-of-thoughtDecomposition (least-to-most)Zero-shot CoT ("think step by step")
04
Communication for the right audience
Calibrate vocabulary and tone for whoever will read it.
12344 lessons
Audience adaptation
05
Guardrails and safety
Refuse, acknowledge, and redirect responsibly.
12344 lessons
Safe refusal (guardrails)

Watching doesn't build the skill. Practicing does.

The difference between this training and a video course:

Video course
×You watch someone prompt
×Feedback: none
×The instructor's generic example
×It ends and you forget
PromptEval Learn
✓You write the prompt
✓A 0-100 score per criterion
✓A real task + commented reference
✓One lesson a day keeps the habit

No made-up techniques

Every lesson teaches a documented technique with the source linked: arXiv papers and the official Anthropic and OpenAI guides.

Wei et al. 2022 (chain-of-thought)Brown et al. 2020 (few-shot)Anthropic prompt docsOpenAI prompting guideThe Prompt Report (2024)

Frequently asked questions

How do you practice prompt engineering?
The practice that works is iterative: write a prompt, see the result, measure what was missing, and refine. On PromptEval Learn you do this with one real task a day and a judge that scores each criterion from 0 to 100, so you know exactly what to improve, not just that it "was vague".
Do I need to pay or create an account to practice?
Today's lesson is free and needs no signup or API key. The full competency tracks, the archive of past lessons, and the commented reference answer are on the Basic plan.
Do I need to know how to code?
No. The lessons are natural-language workplace tasks (support, data, fintech, legal). You write the prompt in English or Portuguese, with no code involved.
How long does it take per day?
About 3 minutes per lesson. The goal is the habit: one technique a day, with a streak, the same way you learn a language.

Your first lesson takes 3 minutes

No signup, no card, nothing to install. Write a prompt and see your score.

Practice today's lesson