About the program
The core challenge is not finding a “magic phrase” that makes a model obey every time. It is turning a task into a clear specification that can be tested. The diploma begins with the concepts users need to reason about language models: probabilistic generation, context, output variability, and knowledge limits. It then builds the prompt itself through goals, context, constraints, output formats, examples, decomposition and iteration. The program then shows when prompting alone is not enough. A task may need retrieved trustworthy context, a structured output, a tool call, or careful handling of documents and multimodal input. A full subject is devoted to evaluation so learners define success before optimization, build test cases, compare versions and analyze failures. The final subject covers privacy, prompt injection, bias, usage rights and human verification, ending with an applied workflow project. Examples are provider- and country-neutral; product details appear only as dated, documented examples.
Practical relevance
The method can be applied to everyday tasks such as summarizing documents, extracting data, drafting content, classifying requests, building a source-grounded assistant or preparing a step that calls a tool. In each case, learners identify what data may be used, write a specification, test the output and decide whether it can be relied on or needs human review.
Why this program
Models and products change quickly, while the skills of specifying a task, controlling context and constraints, verifying outputs and measuring quality are more transferable. The diploma therefore teaches a method that can move across tools, using documentation from several providers only to verify current practices and distinguish stable principles from product-specific details.
Career relevance
The diploma is relevant to people using generative AI in writing, analysis, research, customer service, knowledge management, products, operations or digital tools. It provides practical language for working with data and technology teams and for controlling quality, but it does not promise employment, a professional title or advanced technical specialization.
What you will learn
- Understand the practical foundations of generative AI and language models and their limits.
- Turn a task into a prompt with a clear goal, context, constraints and output format.
- Use examples, decomposition, chaining and iteration when they genuinely help the task.
- Recognize when retrieval, structured outputs or tools are needed instead of simply making a prompt longer.
- Build success criteria and test cases and compare prompt or workflow versions.
- Analyze hallucinations, errors, citations, privacy, prompt injection and bias.
- Design an end-to-end provider-neutral workflow with human verification points and clear documentation.
Learning outcomes (10)
- Explain core generative AI and language-model concepts and how context and probabilistic generation shape outputs.
- Assess generative AI capabilities and limitations, select appropriate tasks, and identify where human verification is needed.
- Design prompts that specify the task, context, constraints, output format, and examples when useful.
- Improve prompts systematically by decomposing complex tasks, testing outputs, and learning from failure patterns.
- Distinguish prompting, grounding/retrieval, structured outputs, and tool calling, and select the appropriate mechanism for a task.
- Build a source-grounded workflow and verify claims and citations before relying on the output.
- Define measurable success criteria and build representative test cases for prompt or workflow quality.
- Analyze quality failures, compare variants, document versions, and monitor the effect of model or configuration changes.
- Identify risks from hallucination, privacy, prompt injection, bias, harmful content, and usage rights, and apply appropriate safeguards.
- Design an end-to-end provider-neutral generative AI workflow with safe inputs, clear instructions, verification, suitable outputs, and reviewable documentation.
Who it is for
- Learners and professionals seeking a structured foundation in generative AI and prompt engineering.
- AI-tool users who want to move from ad-hoc experimentation to testable design.
- Business, product, operations, content, customer-service and knowledge-management practitioners using generative models at work.
- Developers beginning to work with generative models who want foundations in prompting, evaluation and safety before deeper specialization.
Prerequisites
- No advanced AI or programming experience is required.
- Learners should be comfortable reading simplified technical examples, comparing outputs and following structured instructions.
What it does not offer
- It is not a course on training a language model from scratch or advanced deep-learning mathematics.
- It is not a manual for one product and does not guarantee that one prompt formula will behave identically across models and versions.
- It does not grant a professional licence or replace an organization’s approved information-security or privacy policies.
Curriculum
- Subject 1 Open for free
Generative AI and Language Model Foundations
Builds the practical mental model for generative systems, their limits and the role of human verification.
- Subject 2
Prompt Engineering: From Task to Testable Specification
Turns a need into a testable prompt through goals, context, constraints, output formats, examples and systematic iteration.
- Subject 3
Grounding, Tools and Structured Outputs
Explains when work needs retrieved sources, structured outputs or tools and how to control execution and verification.
- Subject 4
Evaluation and Prompt Improvement
Moves improvement from intuition to measurement using success criteria, test sets, rubrics, error analysis and versioning.
- Subject 5
Responsible, Secure and Integrated Application
Combines privacy, security, bias, usage rights and human verification in an integrated applied project.
Subjects unlock in sequence as you progress.
Evidence basis
The blueprint draws on the NIST Generative AI risk profile, the OWASP LLM application risk list, original research on Transformers, few-shot learning and RAG, and current public OpenAI, Anthropic and Google documentation on prompting, evaluation and tools. Vendor documentation is used as dated verification of current practices, not to make the curriculum dependent on one provider.
Reference organizations
- NIST
- OWASP Foundation
- OpenAI
- Anthropic
Frameworks used
- NIST AI RMF: Generative AI Profile — NIST AI 600-1, July 2024
- OWASP Top 10 for Large Language Model Applications — 2025 edition
Methodology
Stable concepts are separated from product details. The sequence moves from a system mental model to prompt specification, grounding/tools, evaluation, and finally safety plus an integrated project.
Program questions
Do I need programming knowledge?
No. The diploma starts at a foundational applied level and most examples can be understood without writing code. Structured outputs and tool calling are introduced conceptually before application; advanced programming remains outside scope.
Is the diploma about ChatGPT only?
No. The curriculum is provider-neutral. Product documentation may appear to illustrate a current practice or implementation difference, but the core skill is specifying the task, context, constraints, output, evaluation and safeguards.
Is there one best prompt formula?
No universal formula is guaranteed. Prompt quality depends on the task, model, context, data and success criteria, so the diploma teaches testing, comparison and systematic iteration.
What is the difference between a prompt and RAG?
A prompt specifies what you want the model to do and how. RAG adds information retrieved from external sources into context when the task requires trustworthy, current or organization-specific knowledge.
Will I learn how to evaluate output quality?
Yes. Subject 4 covers success criteria, test sets, suitable rubrics or metrics, error analysis and controlled comparison of versions before adopting a change.
How does the diploma handle hallucination and privacy?
They are treated as workflow-design concerns: minimize sensitive data, ground outputs when needed, verify claims, define human-review points, and account for prompt injection when external content or tools are involved.
Do I need a paid AI subscription?
The curriculum does not require a specific provider or paid plan. Activities can use an available tool within its policies, and written cases can substitute when a tool is unavailable.
What will I produce by the end?
You will combine the learning into an end-to-end generative workflow design with a clear task, safe inputs, a structured prompt, sources or tools when needed, evaluation criteria, verification steps and documentation.