
Prompt Engineering
Anyone whose job is mostly writing, analysing or deciding
- Length
- 12 weeks
- Modules
- 15
- Per week
- 8 hrs
- Entry
- Computer literacy
What you gain
Works through all 28 prompting techniques on their own job, and becomes the person in the department whose AI work is trusted.
What your organisation gains
Two to four hours a week back per knowledge worker, with output quality measured rather than assumed.
What you build
- A personal AI assistant set up for your own job and writing voice
- An email drafting and reply set for the messages you send most
- A meeting-notes-to-actions workflow
- A question-answering setup over your own documents, with citations
- A reusable prompt library your whole team can work from
Three courses
- 1.1
How language models behave
Weeks 1-3- What the model predicts, and why that explains its behaviour
- Context engineering: what belongs in the window, what to leave out, and why more context is often worse
- Reasoning effort instead of asking for step-by-step working: what the models now do for you
- Sampling controls, determinism, and getting the same answer twice
- Working across English, Amharic, Afaan Oromoo and Tigrinya
- 1.2
Advanced technique
Weeks 4-7- All 28 prompting techniques, worked through against your own job rather than toy examples
- Which of those techniques reasoning models have made redundant, and what replaced them
- Grounding in your own documents, with citation and a refusal path
- Structured output: JSON schema enforcement and constrained decoding
- Tool and function calling, and the Model Context Protocol
- Prompt injection, exfiltration, and keeping personal data out of logs
- Building a personal assistant tuned to your own role and voice
- 1.3
Evaluation and production practice
Weeks 8-11- Golden sets, rubric grading, and LLM-as-judge calibrated against human labels
- Regression suites that block a change on a measured quality drop
- Prompt versioning, review, and A/B testing against live traffic
- Token accounting, caching, and right-sizing the model to the task
- Knowing when prompting is the wrong tool
Capstone, week 12
One real task from your own work, with a 30-example evaluation set and measured before-and-after accuracy.



