Capstone: Ship it
The launch checklist, the demo with a deliberate flaw, the rubric, and the Certified AI App Builder — Web credential — plus where the shipping road goes from here.
The capstone is the shipped thing itself, presented like a builder: live URL, honest demo, real numbers, next steps. The final checklist before you call it done:
- 1The launch checklist, all green: live URL on your domain (or a deliberate decision not to buy one); critical path passing against production; env vars verified against
.env.example; spending cap, rate limit, and billing alert confirmed live; access decision implemented; privacy note linked; README complete enough that a stranger could run it locally in ten minutes; repo history clean of secrets (run the search one last time). - 2Record the 5-minute demo: the loop end to end on the live URL — paste a message, watch it arrive, open it, draft with AI (let the stream show), edit, send, status flip, refresh to prove persistence. Then — the move every capstone in this academy requires — show one flaw on purpose: trip the rate limit, or show the
[NEEDS INFO]escape firing, or the empty state, and explain the design behind it. Builders who show their systems' edges are builders who know their systems. - 3Write the ship report (one page): what you built and for whom; the stack in one sentence you fully understand; three hardest bugs and what each taught you; the cut list as your v2 roadmap, honestly prioritized; and your AI-assistance reflection — where the assistant excelled, where it misled you, and what you now review hardest in generated code. (That reflection is the AI-appendix discipline from the analyst course, applied to code — and it's the section reviewers read first.)
- 4Submit: URL + repo + demo + ship report.
Certification rubric
- It ships and it works (35%) — live, persistent, the loop clean end to end, AI feature streaming with failure paths that actually catch.
- It's built right for its size (30%) — three-place separation clean, secrets never client-side or in history, doors validated, rate limit live, tests exist and pass, spec and code agree.
- It's ready for humans (20%) — polish pass evident, accessibility floor met, states designed, privacy note honest, access decision deliberate.
- The builder understands it (15%) — the ship report demonstrates comprehension: bugs explained causally, AI reflection specific, v2 priorities defensible. The credential certifies you, not the assistant.
Passing earns the Certified AI App Builder — Web credential (ID format EDOVA-WEB-2026-XXXX, independently verifiable at edova.ai/verify): you took an idea to a specified, built, hardened, polished, deployed product — the full loop, which most people who 'know how to code' have never once completed.
Four roads, by appetite: Ship an iOS App with AI to do it again on the other platform (you'll be startled how much transfers); Building Customer-Facing AI Agents if Replyable's draft feature made you want to build the conversation itself; Prompt Engineering & Context Design to turn your lib/prompts.ts instincts into production-grade craft; and eventually LLMOps, when an AI feature you shipped has real users and 'is it still good?' becomes a daily question. And v2 of Replyable is sitting right there in your cut list — auth first; you've earned the complexity now.