Prompt Engineering to Context Engineering
Learners who need AI to work consistently with an organisation's documents, terminology and standards.
What you will be able to do
- Inventory the knowledge assets a recurring AI task actually depends on.
- Structure documents for retrieval, reuse and clear ownership.
- Write system instructions that reference approved context rather than general knowledge.
- Test whether a knowledge base improves answer quality, and prove it.
- Set maintenance rules for outdated or conflicting content.
How this is assessed
Reusable knowledge base an AI can work fromSubmit a structured, sourced knowledge set for one real recurring task, together with the system instructions that sit above it, your test prompts with results, and your maintenance notes. It must be built on genuine material — invented documents cannot be assessed, because the whole exercise is about the mess real material comes in.
| Grade | What earns it |
|---|---|
| Pass | A usable knowledge base with instructions and evidence that it was tested. |
| Merit | Retrieval quality demonstrated against the test set, with results a reader can check. |
| Distinction | Strong source governance and deliberate gap handling — the system knows what it does not know, and says so. |
Before you submit
- The task and the people who do it are described in a short paragraph.
- Every knowledge chunk carries a source, an effective date and a named owner.
- System instructions state role, source hierarchy and what to do when the answer is absent.
- At least twenty test prompts, including four the base should decline to answer.
- Results are recorded separately for retrieval, accuracy and abstention.
- Maintenance notes name who updates it, how often, and how a user reports a wrong answer.
Handing it in
Marked by a knowledge-management, operations or AI implementation professional who has prepared organisational content for AI-assisted work. You will need an account to submit, so that the mark has somewhere to land. Sign in or create one. The modules are open either way.
Modules
6 modules, self-paced. Nothing is timed and nothing is scored.
- 1The context gapWhere clever prompting stops working.
- 2Content auditFind the authoritative files and name their owners.
- 3Knowledge designChunk, label and version the material worth keeping.
- 4The instruction layerRole, source hierarchy and limits.
- 5EvaluationCoverage, accuracy and abstention.
- 6HandoverUpdates, governance and what happens when it drifts.
This course stacks
Can take a real business process, give an AI the context it needs, and automate it end to end without losing accountability.
A credential is issued when all of its artifacts have passed. It records the courses, the bands and the dates, nothing else, and no score.