Capstone: Ship a Real Full-Stack AI-Assisted Project

The Real Test

This capstone is not a toy. You will ship a full-stack AI-assisted project from empty folder to live URL, with tests, CI, deploy, and a short technical write-up. The rubric rewards depth, judgment, and process — not just a working demo.

Project options (pick one)

• Habit tracker with analytics: auth, CRUD, charts, streaks, export
• AI-powered study buddy: upload notes, generate quizzes, track mastery
• Team standup bot: Slack integration, summarizes yesterday/today/blockers, stores history
• Expense dashboard: receipt upload, OCR, categorization, monthly report
• Open-source contribution: pick a real OSS project, ship one meaningful PR

Non-negotiable requirements

1. Full stack: frontend + backend + database
2. Authentication (not demo-grade — real sessions or JWTs)
3. Automated tests with coverage >= 60% on business logic
4. CI/CD: every push runs tests; main branch deploys automatically
5. Deployed to a public URL (Vercel, Fly.io, Railway, Render — anywhere real)
6. README with: problem statement, architecture diagram, setup steps, deploy steps
7. Git history with at least 15 commits and descriptive messages
8. A 5-minute video walkthrough (screen recording + narration)

The agentic workflow to follow

Day 1 — Plan Use Claude Code or Codex CLI to create a CLAUDE.md / AGENTS.md with: - Stack choices and reasons - Schema and API sketch - First 10 tasks in priority order Open a GitHub repo, push the plan as your first commit. Day 2-3 — Scaffold Agent scaffolds the backend (routes, DB, auth). You review every file before committing. Write tests for each route as you go. Day 4-5 — Frontend Prototype UI with v0 or Lovable, export to repo. Or build in Cursor using shadcn/ui components. Wire frontend to backend, end-to-end. Day 6 — CI/CD Set up GitHub Actions with test + lint + build. Connect deploy previews (Vercel is easiest). Add an AI release-notes action to every PR. Day 7 — Polish and Record Fix edge cases, improve error messages. Write the README. Record the video walkthrough.Seven focused days. Not seven intense days — regular days, with a plan.

The three reviews that matter

The write-up that gets you hired

Your README or accompanying blog post should answer: what did the agent do well? Where did it fail and how did you recover? What did you design that it could not have generated? Interviewers love this because it signals self-awareness, and self-awareness is the rarest signal.

Grading rubric

• Working deployment: pass/fail
• Test coverage and CI green: pass/fail
• Code quality: readable, typed, no obvious security holes
• Judgment: can you articulate tradeoffs in the video?
• Process: git history shows thoughtful iteration, not one giant commit
• Documentation: someone could set up your project in 10 minutes

After the capstone

1. Post the repo publicly with a strong README
2. Share the demo link on LinkedIn with a before/after of what you learned
3. Tag 3 engineers you admire and ask for honest feedback
4. Pick your next project — the skill compounds only if you keep shipping
5. Consider contributing back to an open-source tool you used in the build

The future belongs to engineers who ship. AI just made shipping more accessible than ever.

— A Tendril alum

The big idea: you now have a complete AI-assisted engineering workflow — from planning with an agent through testing, deploying, and documenting. That's the baseline of an AI-literate engineer in 2026. Go build something that matters.