The capstone is not an exam. It is a guided journey through the complete AI product lifecycle. By the end, you will have a shipped product, a postmortem, and a portfolio piece that demonstrates end-to-end mastery.
The best way to learn product development is to do product development. This capstone gives you structure, templates, and feedback loops so your learning is as efficient as it is comprehensive.
The AI Product Handbook31.1.1 What the Capstone Covers
The capstone simulates a real AI product development engagement from kickoff to postmortem. You will work through nine phases that mirror the actual lifecycle of an AI product at a modern company:
The capstone simulates a real AI product development engagement that spans twelve weeks, organized into nine distinct phases.
Phase 1 covers Use Case Identification during weeks one and two, establishing the problem space and opportunity. Phase 2 focuses on Discovery and Research in week three, diving deep into user needs and market context. Phase 3 introduces AI-Native Design in week four, where teams develop trust-centered and explanation-focused user experiences. Phase 4 brings the Prototype to life across weeks five and six through vibe coding and rapid iteration.
Phase 5 establishes the Eval Suite in week seven, creating the success criteria and automated tests that will govern quality. Phase 6 tackles System Architecture during weeks eight and nine, designing production-ready infrastructure. Phase 7 develops the Rollout Plan in week ten, preparing for launch with monitoring and rollback procedures. Phase 8 addresses Governance and Trust Planning in week eleven, establishing policies and compliance frameworks. Phase 9 builds the Metrics Dashboard across weeks eleven and twelve, creating the dashboards that will guide ongoing decision-making. A final Phase 10 for Reflection and Postmortem completes week twelve, ensuring learning is captured for future endeavors.
31.1.2 Suggested Timeline
The 12-week timeline is designed for an intensive but sustainable pace. Adjust for your context. The first two weeks focus on Use Case Identification, where teams engage in problem discovery, stakeholder alignment, and AI applicability assessment, ultimately producing an opportunity brief and problem statement. Week three centers on Discovery and Research, conducting user research, workflow analysis, and competitive landscape review to generate research synthesis and user personas.
Week four introduces AI-Native Design, covering UX design with trust and explanation considerations plus fallback design, resulting in a design spec and success metrics. Weeks five and six bring the Prototype to life through vibe coding, rapid prototyping, and user testing, yielding a functional prototype and test feedback. Week seven establishes the Eval Suite through test dataset creation, baseline metrics, and success criteria definition, producing an eval pipeline with passing tests. Weeks eight and nine tackle System Architecture, designing reference architecture, component design, and scalability planning, resulting in an architecture document and security review.
Week ten develops the Rollout Plan including launch strategy, monitoring, support, and rollback procedures, producing a launch checklist and rollback plan. Week eleven addresses Governance and Trust Planning through policy development, compliance work, and audit trail setup, yielding a governance document and compliance checklist. Weeks eleven and twelve also build the Metrics Dashboard through metrics selection, dashboard building, and alert threshold configuration, producing a live dashboard and review cadence. Finally, week twelve conducts the Postmortem, examining what worked, what did not, and lessons learned to create a postmortem document with recommendations.
31.1.3 Team Structure Recommendations
The capstone can be done individually or in teams of 2-4. Each structure has tradeoffs:
Solo work offers full ownership of all decisions and works best for deep personal learning, requiring a time commitment of ten to fifteen hours per week over the twelve weeks. Teams of two allow participants to pair up on product manager and engineer roles, splitting phases by strength, with each person investing six to ten hours weekly. Teams of three to four enable role assignment across product management, design, engineering, and data functions, providing the best simulation of real product teams and requiring a more manageable five to eight hours per person weekly. If working in teams, rotating roles across phases ensures everyone experiences each dimension of AI product development.
31.1.4 Success Criteria
A capstone project is considered successful when it meets the following criteria:
A capstone project is considered successful when problem validation has been achieved through user research that confirms a real problem exists rather than an assumed solution. The solution must demonstrate genuine AI appropriateness by leveraging AI capabilities in ways that would be difficult or impossible without AI. Teams must deliver a functional prototype that demonstrates the core value proposition effectively. Eval coverage ensures automated evaluations catch regressions and validate quality consistently. The production architecture represents a system design capable of scaling with security and privacy considerations addressed comprehensively. Launch readiness requires a rollout plan with monitoring, rollback procedures, and support processes in place. A governance framework with policies and audit trails supports compliance and trust. Finally, an honest postmortem provides a candid retrospective that identifies what you would do differently.
31.1.5 The Five Running Examples
Throughout this chapter, we will reference five hypothetical AI products at various stages. These running examples illustrate how teams navigate each phase:
Throughout this chapter, we will reference five hypothetical AI products at various stages that illustrate how teams navigate each phase. QueryWise is an AI-powered search assistant for enterprise knowledge bases that uses RAG to answer natural language questions about internal documents, serving as an example that maps to phases one through three covering discovery and design. DraftBot is an AI writing companion for marketing teams that generates first drafts and learns brand voice over time, representing phases four and five focused on prototype and evaluation. SupportFlow is an AI customer support system that routes tickets and suggests responses, covering phases six and seven on architecture and rollout. FraudFinder is a real-time fraud detection system for financial transactions, illustrating phases eight and nine dealing with governance and metrics. HealthCoach is a personalized health coaching application using multimodal AI, mapped to phase ten for postmortem and reflection.
31.1.6 Capstone Assessment Rubric
Your capstone will be evaluated on four dimensions, each scored on a 1-4 scale. Problem Discovery measures whether you identified a genuine user problem through research. A score of 1 indicates no clear problem or user research was conducted, while a 2 means the problem was stated but not validated with actual users. Scoring a 3 requires problem validation through research, and a 4 demands deep user insight with strong validation evidence. AI Integration evaluates how meaningfully AI contributes to your solution. A 1 suggests AI was used as a gimmick with no real value, a 2 shows AI was applied superficially, a 3 indicates AI is core to the value proposition, and a 4 demonstrates AI is essential and exceptionally well-executed. Engineering Quality assesses your technical implementation. A 1 means the code does not run, a 2 indicates it runs but lacks eval coverage, a 3 shows it works with basic evals, and a 4 confirms production-ready quality with comprehensive evaluations. Thoughtfulness examines your reflective practice. A 1 suggests a superficial postmortem, a 2 shows some reflection is present, a 3 indicates candid analysis of failures, and a 4 demonstrates deep insights with actionable next steps.
31.1.7 Getting Started
Before beginning Phase 1, ensure you have the following in place:
Before beginning Phase 1, ensure you have access to AI APIs by obtaining API keys for at least one LLM provider such as OpenAI, Anthropic, or Google. Your development environment should be ready with an IDE or editor configured and version control established. Prototyping tools should be accessible, including vibe coding platforms like Cursor or Copilot. Eval infrastructure requires a test framework such as pytest or Jest to be prepared. Documentation setup provides a place to capture research, decisions, and learnings throughout the process. If working in teams, stakeholder availability means clear roles and communication channels should be established.