Specialist Preparation Program in Generative AI and Prompt Engineering is a free program on ARPS, the Arab Platform for Professional Studies. An applied program that prepares specialists to specify, prompt, test and govern generative-AI work with explicit controls. The program progresses from task specification and model limits to bilingual prompt and output engineering, retrieval and grounding, quantitative evaluation and operational.
Optional certificate fee 50% launch discount
$90 instead of $180
331 AED or 338 SAR
The program progresses from task specification and model limits to bilingual prompt and output engineering, retrieval and grounding, quantitative evaluation and operational trials, then privacy, security, rollout and handoff. Every lesson teaches a distinct skill through a worked case, self-check exercise and feedback. Good wording is never treated as sufficient; each result is tied to evidence, an acceptance rule and an authority boundary.
Learners produce task cards, output contracts, test cases, trial plans and reviewable rollout packages.
Moving a generative model from an attractive demo into dependable work requires specifications, data, tests and governance rather than a collection of prompt phrases.
Professional practice begins with a task card that identifies the user, decision, permitted input and acceptable output, while limiting success claims to what was actually tested. Learners distinguish deterministic rules, retrieval and generation, selecting the simplest path that fits the evidence rather than the newest tool. The lessons explain model limits and context windows, including why text being present does not guarantee correct use. Through worked cases and a checkable exercise, learners produce a specification that separates what may be generated, what must stay grounded and what must be reported as unknown. The module output is a testable task card and authority boundary ready for review before prompt writing.
This module converts a task specification into an ordered prompt containing the task, context, rules, examples, output contract and authority limits. Learners classify messages, construct structured outputs, and verify field types and meaning instead of treating syntactically valid JSON as sufficient. In bilingual summarization they preserve numbers, negation, conditions and certainty while separating editorial translation from adding facts. Larger tasks are divided into stages with handoff contracts that expose the location of failure. Worked examples and exercises culminate in a testable prompt and output contract, including boundary cases and focused retesting after change. The result can be inspected by another practitioner rather than accepted because it sounds fluent.
The module builds a retrieval path that starts with a source owner, effective status, date and scope, then chunks material without separating a rule from its exception. Access controls are applied before text reaches the model. Learners design an answer contract that requires grounding in retrieved passages and exposes conflict or missing evidence instead of guessing. When an answer fails, they trace the chain from source and indexing through candidates, selected context and writer behavior, separating search failure from negation reversal or meaning loss. Worked cases and exercises require locating the faulty layer and choosing a suitable regression test. The module output is a documented diagnosis and bounded grounded answer, not a cosmetic prompt rewrite.
This module turns quality from an impression into a test set representing decisions, boundary cases and expected failures, with defensible reference answers and a scoring rubric. Learners preserve held-out cases whose keys were not exposed during development and compare two versions on identical inputs under controlled settings. They calculate precision, recall, percentage points, relative improvement and cost per accepted output using declared denominators, while preventing an average from hiding a critical failure. The comparison expands into an operational trial that records attempts, rework, latency and cost and matches the intended production retry policy. Learners produce a numerical, gate-based decision for limited rollout, repair or rollback rather than choosing the most persuasive-looking response.
The module integrates privacy, security and rollout controls into a practical path that begins by recording purpose, data source, usage rights, minimization and synthetic alternatives. Learners distinguish permission to open a file from permission to send it to an external service, and treat retrieved content as untrusted data that cannot grant itself authority. They apply least privilege, reauthorization by the executing system, approval, receipts and request identifiers for high-impact effects. The final lessons design a limited rollout that monitors quality, cost and incidents with stop and rollback thresholds. The module output is a handoff package linking version, sources, tests, owners and an impact map so future changes trigger the correct review.
Subjects unlock in sequence as you progress.
Bounded technical claims use primary research and official sources; applied procedures are identified as instructional synthesis.
Primary sources for claims, followed by synthetic cases and educational applications with stated limits.
Assessments follow the program blueprint and may use module assessments, a final assessment, or a combination, under the approved attempt policy.
Assessment attempts: 2 per assessment, 7 days apart
The pass mark is announced before the program is published.
The credential confirms completion and successful assessment; it does not represent professional licensure unless explicitly and lawfully stated.
Every certificate has a public verification code. Verify a certificate
The real design with fictional details; opens in a new window.
Early lessons require no programming and explain contracts and metrics in practical language; basic technical literacy helps with examples.
No. It teaches testing, boundaries and rollout decisions for a defined use case.
Each uses a complete synthetic case, asks for a decision or specification, and provides reasoned feedback for comparison.
Yes. Lessons separate source preparation, retrieval and context selection from grounded answer generation.
Every lesson has a knowledge check and every module has an independent-bank assessment; the pass threshold remains a central policy.
Real data should not be submitted until purpose, rights, controls and organizational approval are established; exercises do not require it.
Yes. Specialist Preparation Program in Generative AI and Prompt Engineering is free on ARPS: every lesson, exercise and assessment is free, the first subject is open to read without an account, and the program is available in Arabic and English. Issuing a certificate is optional; it is a training completion certificate showing you passed the program's assessments, not a university degree, a professional licence or an endorsement by an external body.
The subject of this program is also known as: Prompt Engineering, Generative AI, GenAI Prompting, Prompt Design. Learners usually find it by searching for: free prompt engineering course, learn generative AI online, writing effective prompts, testing generative AI outputs, prompt engineering for work tasks.
Programs from the same field or route that may complement what you learn here.
Data, AI, Digital, IT & Cybersecurity
A foundational, applied diploma on how generative AI systems work at a practical level, how a need becomes a clear testable prompt, and how prompts fit into grounded workflows, structured outputs and tools, with quality evaluation, privacy, security and human verification.
Data, AI, Digital, IT & Cybersecurity
An advanced program for designing, securing, evaluating, and operating agentic workflows.
Data, AI, Digital, IT & Cybersecurity
Learn to prepare an initial cloud decision for a small workload using demand, capacity, responsibility, access, recovery and cost evidence. Move from classifying an offer to testing booking, backup, monitoring and migration, then hand over a record of what was shown and what needs owner approval. Cases are fictional and do not grant production authority or professional certification.
Data, AI, Digital, IT & Cybersecurity
Turn software requirements into checkable acceptance conditions and reproducible tests. Record execution and defects, then write an evidence bounded release recommendation. The program covers design, risk, accessibility, authorization, performance and automation.
