Formal training programme · Handbook v2.0
How to decide, specify, test and launch products built on artificial intelligence
This programme teaches the decisions that make or break products built on artificial intelligence (AI): whether to use AI at all, how to specify behaviour that varies, how to test it, how to pay for it, how to govern it, and how to defend a launch decision. It follows the structure of leading public courses and runs every assignment on one case company, a fictional Singapore polyclinic group, so that learners from any background work on the same realistic problems.
Conventions. Every acronym is written out in full the first time it appears on a page, with the short form in brackets. The glossary at the end lists them all.
The programme suits anyone who helps decide what an AI product should do, or who wants to move into AI product work. You do not need to write code. Each module has an optional technical lab for those who want hands-on practice.
| Your background | What you already bring | Where to put extra effort | Modules that will stretch you |
|---|---|---|---|
| Business analyst | Requirements, process mapping, stakeholder work | Requirements for behaviour that varies; measuring quality | 2, 5, 6, 9 |
| Data analyst | Metrics, data quality, experiments | User discovery, user experience, strategy | 3, 7, 11 |
| Software engineer | Architecture, testing, delivery | Problem framing, pricing, governance | 1, 3, 9, 10 |
| Designer or researcher | User research, interaction design | Technical trade-offs and evaluation | 2, 5, 6, 8 |
| Domain expert (clinician, consultant, operations lead) | Deep knowledge of the work and its risks | Technical fluency, requirements, economics | 2, 4, 6, 9 |
| Product manager moving into AI | Discovery, prioritisation, launch | Technical fluency, evaluation, governance | 2, 5, 6, 10 |
The Week 0 questionnaire suggests a starting emphasis for you, using this table and your own self-ratings against the twelve modules.
Case company · fictional
Harbourview is a fictional group of public polyclinics (primary care clinics) in Singapore. It offers general practice, chronic disease care, vaccinations, health screening and an on-site pharmacy at each clinic. Patients book, check queue status and see their medicines in the Harbourview app. Any resemblance to a real organisation is unintended.
The board has asked the new head of digital products to build an AI portfolio within 18 months, under a tight budget and close scrutiny from the Ministry of Health. Five initiatives are on the table. Each module uses one or more of them.
| Initiative | What it would do | Kind of AI | Main tension |
|---|---|---|---|
| A. Medicine questions assistant | Answers patients' questions about their own prescribed medicines in the app | Large language model (LLM) with retrieval from approved leaflets | Patient safety versus convenience |
| B. Booking and symptom-routing chatbot | Helps patients book the right service and flags urgent symptoms for a nurse | LLM with rules and human handoff | Catching emergencies versus over-referral |
| C. Consultation note drafter | Drafts the doctor's visit notes from the recorded consultation | Speech-to-text plus LLM | Doctor time saved versus errors in the medical record |
| D. Missed-appointment predictor | Predicts who will miss an appointment so staff can send reminders or overbook | Classical machine learning (ML) | Efficiency versus fairness to some patient groups |
| E. Patient feedback analyser | Groups thousands of free-text comments into themes for clinic managers | LLM classification | Speed versus missing serious complaints |
Each module adds a case pack with the extra material needed: interview notes, sample outputs, cost figures, policy extracts. If you are already working on an AI product of your own, you may use it for any assignment instead, as long as it raises the same questions.
Every module has two kinds of assessment. Formative assessment checks understanding while you learn. It gives instant feedback, can be retaken, and does not count toward your grade. Summative assessment at the end of each module measures what you can do once the module is over, and it is graded.
| Component | When | Format | Feedback and threshold |
|---|---|---|---|
| Formative Checkpoint quiz A | After the lesson | 10 Yes/No questions and 3 short-answer questions on the core concepts | Yes/No answers are checked instantly, with an explanation. Short answers are compared with a model answer. Mastery is 8 of 10 Yes/No correct; retake until reached. |
| Formative Checkpoint quiz B | After the case study | 5 Yes/No and 3 short-answer questions applying the concepts to Harbourview | As quiz A. Mastery is 4 of 5. |
| Component | Weight | Format | How it is marked |
|---|---|---|---|
| Summative Module deliverable | 60% | The written work product set in the module, for Harbourview | Rubric in Section 6, four criteria |
| Summative End-of-module test | 25% | 10 Yes/No and 5 short-answer questions, closed book, 30 minutes, taken once | Yes/No: 1 mark each. Short answers: 0–3 marks each against a marking guide. Total 25 marks. |
| Summative Viva | 15% | 10 minutes of oral questioning (a viva voce) on your deliverable, with an assessor or a peer using the viva guide | Can you defend your choices and handle a challenge you did not prepare for? |
| Step | Time | What happens | Assessment |
|---|---|---|---|
| 1. Lesson | 60–90 min | Concepts with worked examples from real AI products | Quiz A |
| 2. Case study | 45 min | The module's Harbourview case pack | Quiz B |
| 3. Assignment | 3–4 hrs | You produce the module deliverable | Deliverable |
| 4. Close | 60 min | End-of-module test, viva and feedback | Test Viva |
| Optional lab | 1–2 hrs | Hands-on technical exercise for those who want it | Not assessed |
Term I · Foundations Term II · Mastery Term III · Strategy and leadership Capstone
| Week | Module | Harbourview initiative | Summative deliverable |
|---|---|---|---|
| Week 0 | Orientation, self-assessment and pre-test Available | A | Baseline memo, recorded but not graded |
How to start. Sign in with your email address; there is no password. Your course page then takes you through Week 0 and each module in order. The programme is self-paced: each module opens once you have reached mastery on the previous module's checkpoint quizzes and emailed its assignment.
Each criterion is scored 0–4 and weighted equally. The total is scaled to the 60% deliverable weight.
| Criterion | 4 · Distinction | 3 · Merit | 2 · Pass |
|---|---|---|---|
| Decision quality | Clear recommendation; alternatives and trade-offs quantified | Clear recommendation; alternatives named | Recommendation present but weakly argued |
| Evidence | Case data, evaluation results or cost models drive the conclusion; assumptions labelled | Evidence present but partly assumed | Mostly assertion |
| Risk handling | Failure modes, who is harmed, mitigations and remaining risk named | Main risks named and mitigated | Risks listed only |
| Communication | An executive can act on it after five minutes | Clear but long | Hard to follow |
A score of 0–1 on any criterion means the deliverable goes back for a rewrite before it is graded.
You sign in with your email address. Each time, the site sends a one-time code to your inbox, so there is no password to create, remember or leak. The programme stores only two things: your email address and your quiz and assignment progress. It does not ask for your name, employer or anything else. You can ask for your account and progress to be deleted at any time. A full privacy notice, written to meet Singapore's Personal Data Protection Act (PDPA), will be published before the programme opens.
| Source programme | Structure | Used in |
|---|---|---|
| Duke University, AI Product Management Specialization (Coursera) | Three courses: ML foundations for product managers; managing ML projects; human factors in AI | Modules 2, 3, 7, 10 |
| AI Product Academy (Marily Nika), AI Product Management Bootcamp (Maven) | An introductory course, an advanced course on strategy, organisation and ethics, and a capstone where you ship a product | Modules 1, 3, 8, 12 |
| Hamel Husain and Shreya Shankar, AI Evals for Engineers and Product Managers (Maven) | 8 lessons on error analysis, synthetic data, using an LLM as a judge, debugging retrieval, continuous integration and red-teaming | Modules 4, 5, 6 |
| Reforge, AI Foundations, AI Evals, AI Prototyping, AI Strategy, AI Growth | Foundations, Mastery and Strategy pillars taught by practitioners | Term sequence; Modules 8, 9, 11 |
| Product School, AI Product Certification | 6 modules, about 15 hours over 3–4 weeks, ending in an AI requirements document | Module 4 |
| Short form | In full |
|---|---|
| AI | Artificial intelligence |
| ADR | Architecture decision record: a short document recording a design choice, the options considered and why one was chosen |
| Eval | Evaluation: a repeatable test of whether an AI system's outputs are good enough |
| EU | European Union |
| IMDA | Infocomm Media Development Authority, Singapore |
| ISO/IEC 42001 | International Organization for Standardization and International Electrotechnical Commission standard for AI management systems |
| LLM | Large language model, such as Claude or GPT |
| ML | Machine learning |
| NIST AI RMF | United States National Institute of Standards and Technology, AI Risk Management Framework |
| OWASP | Open Worldwide Application Security Project, which publishes lists of common security risks |
| PAIR | People + AI Research, a Google team that publishes design guidance |
| PDPA | Personal Data Protection Act, Singapore |
| POC | Proof of concept: a short, limited trial to show a product works |
| PRD | Product requirements document |
| TCO | Total cost of ownership: everything an option costs over a fixed period, including people's time, price rises and the cost of leaving |
| RAG | Retrieval-augmented generation: the model is given relevant documents found by a search step before it answers |
| UX | User experience |
Handbook v2.0 · 20 October 2026 · Rewritten for a general audience with the Harbourview Polyclinics case. Week 0 revised 21 October 2026 once all twelve modules were published.