Responsible AI in Product Development: From Idea to Launch

This course contains the use of artificial intelligence. Your team is about to ship an AI feature. Someone asks whether it is fair, whether users will know they are talking to an AI, who can override it, and what the EU AI Act or a US state law expects from it. Those answers cannot wait for the week before launch. They have to be built into discovery, requirements, design, testing and release, the same way security and quality are. In this course we follow a fictional HR-tech company, Hollinsway, as its product team takes two AI features to launch: Shortlist, which ranks job applicants for recruiters, and Ask Hollinsway, a generative assistant for recruiters and candidates. We work alongside its product manager, its machine learning engineer and its product designer, and at every stage we produce one part of a responsible AI product pack that you can reuse at work. We start by asking whether a feature should use AI at all, who it affects, and how risky it is under the EU AI Act and US state laws. We run a harms-modelling workshop and complete an impact assessment. Then we turn principles and regulation into user stories with acceptance criteria: disclosure notices, human oversight, logging, appeal paths and data checks. We design the experience with Google's People + AI Guidebook and Microsoft's Guidelines for Human-AI Interaction, and we make it accessible under WCAG 2.2. Then we test, with free tools on our own computer. We produce a fairness report with Fairlearn and AIF360 on a synthetic hiring dataset, apply a mitigation and write the trade-off memo that a product owner signs. We evaluate the generative assistant with promptfoo against a small open-weight model running locally, find a real failure and fix it, then add misuse and prompt-injection cases. We audit an AI disclosure panel with axe-core and fix what it finds. We write a model card, run a launch-gate review and plan the rollout. Finally we set up monitoring, feedback and appeal loops, and a product-side response for when something goes wrong. Every lab ends with the decision the product manager makes from its output, so you can follow the course without writing code, and engineers on your team can run every command. The frameworks are the ones product teams are asked about: the NIST AI Risk Management Framework and its Generative AI Profile, the OECD AI Principles, Microsoft's Responsible AI Standard, and ISO/IEC 42001 and 42005 where they meet product work. Regulation is taught as product requirements, with the dates as they stand when the course is recorded. Your instructor is a practitioner who helps product teams build and govern AI features, not a lawyer. This course is not legal advice: check your own obligations with your own counsel. What you get: 44 lectures, about 4.6 hours, including 11 hands-on labs built from real tool output Section quizzes, four assignments with model answers, a capstone and a 50-question practice test Downloads: a risk-triage worksheet, an impact assessment template, a requirements library, UX pattern cards, a test plan template, model card templates, a launch-gate checklist, a monitoring plan and the lab kit All companies, people and data in the course are fictional. Real cases and enforcement actions are cited from their primary sources.