2–4 person teams
Groups of 3 work best. Roles are soft — every member commits code. No “non-technical” seats.
Pick one track
Campus Life, Personal Growth, or Creative Tools. The team may combine elements but the spine should be one theme.
Demo Day at Week 12
Five-minute live demo, three-minute Q&A. Public — invited employers, alumni, career-services team.
The three tracks
Campus Life
Products that serve daily university life. Low friction to get classmates to try it — real user feedback within 4 weeks. Examples: lecture-note swap with AI summaries, roommate matcher, campus event radar with a personalised weekly digest, dorm chore-and-bill splitter.
Personal Growth
Products where the user’s state matters — logs, history, streaks. Teaches teams about private data, consent, and longitudinal UX. Examples: AI habit coach with weekly LLM-authored reviews, reading companion with spaced repetition, journaling assistant with monthly reflections, workout planner with volume-progression tracking.
Creative Tools
Generative products with a clear loop: input → AI output → user iterates. Most visually demo-able at the showcase. Examples: AI short-story studio with branch/merge narrative choices, music-prompt playground with shareable rooms, comic-panel generator with PDF export, resume-video script maker.
Project requirements (non-negotiable)
Every capstone, regardless of track, must include:- Authentication — multiple real human users, not a demo account. Use Neon Auth or equivalent.
- Persistent state per user — Neon Postgres as primary store; per-user data isolation verified.
- At least one AI feature in the critical path — removing the AI should break the core value. No “AI-washing”.
- Real-time notification — Slack or email, triggered by a user action.
- Deployment to Vercel on a shared team repo. Merge via pull request; no direct-to-main.
- A one-page landing page explaining the product to a stranger in under 30 seconds — hero, problem, demo video, sign-up.
- A working
/feedbackform that persists to Neon and pings the team’s Slack.
Default stack
Unless a track strongly demands otherwise, teams ship on the same stack you learned across Weeks 1–8:| Layer | Default |
|---|---|
| Framework | Next.js 14 App Router + Tailwind |
| Hosting | Vercel |
| Database + ORM | Neon Postgres + Drizzle |
| Auth | Neon Auth (Google + GitHub providers) |
| File storage | Vercel Blob |
| LLM in product | Gemini 2.5 Flash for cheap calls; Claude for reasoning / agents |
| Notifications | Slack Incoming Webhooks or Resend for email |
| PDF / docs | Typst via skill (where applicable) |
Timeline
| Week | Focus |
|---|---|
| Week 9 | Kickoff, scope, repo. By Day 3: core data model + auth + one end-to-end vertical slice deployed. |
| Week 10 | Build the critical path. AI-in-the-critical-path feature working end-to-end. First external user test logged in RESEARCH.md. |
| Week 11 | Harden, polish, market. Landing page, onboarding, accessibility pass, five real users signed up. |
| Week 12 | Demo Day. 5-minute live demo per team, 3-minute Q&A with instructor + invited guests. |
Demo Day requirements
At Demo Day every team presents in this order, in roughly five minutes:- Problem statement — 30 seconds. Who hurts? How often?
- Solution — 30 seconds. The product in one sentence.
- Tech stack — 30 seconds. What from Weeks 1–8 did you reuse; what’s new.
- Live demo — 3 minutes. Sign up a fresh user on stage; walk through the critical path; land the notification moment.
- Challenges faced — 1 minute. One technical, one team, one user.
- What’s next — 30 seconds. Honest: is this worth keeping alive past Week 12?
Evaluation
| Dimension | Weight | What “excellent” looks like |
|---|---|---|
| Functionality | 30% | Every promised feature works end-to-end on the live site during the demo. Real external users ran the critical path at least 5 times pre-demo. |
| AI in critical path | 20% | The AI feature cannot be removed without the product losing core value. Thoughtful prompt design, visible in the repo as files. |
| Code quality + collaboration | 15% | Healthy commit history; every member shipped real features; PR reviews happened; no secrets in git. |
| UI / UX | 15% | Mobile works. Empty states exist. Error messages are in human language. Onboarding surfaces value in under 2 minutes. |
| Deployment + reliability | 10% | No 500s during demo. Vercel logs are clean. Secrets only in env vars. Migrations documented and reversible. |
| Presentation | 10% | All team members speak. Problem lands in the first 30 seconds. Demo is rehearsed. Post-mortem is candid, not marketing. |
Deliverables checklist
- Live URL (Vercel production domain).
- Public GitHub team repo with README + setup instructions.
- Landing page with problem + solution + demo video.
- 2-minute demo recording (unlisted YouTube / Vimeo / self-hosted fine).
- 1-page post-mortem (what went well, what hurt, what you’d cut).
- List of five external users (with permission to cite first names).
- Five-minute live demo on Demo Day.
Prerequisites: Complete Week 8 — Typst PDF Automation first. The capstone assumes you own the Week-1-to-8 stack and can extend it. If you’ve never shipped a Vercel deploy through AI before, start at Week 1.
Download Capstone PDF
Full spec · 604 KB — Track examples, week-by-week timeline, full evaluation rubric, instructor guidance.
Back to series overview
Build a Full-Stack AI Site