Shipyard AI

Learn anything with AI. Prove it by shipping.

An AI engineering studio for college students. Three weeks. Five students. One working practitioner. Real projects.

The problem we exist for

You're a CS major. You work hard, your grades are fine — and you still can't land an internship. Every posting asks for AI/ML experience you were never given a way to get. The uncomfortable truth:

  • The curriculum lags by years. Syllabi are revised by committee on academic timelines; the AI industry ships monthly. What you're tested on and what employers hire for have quietly diverged.
  • The projects aren't real. Coursework hands you toy problems with known answers — built once, graded, deleted. No real user, no real data, no measured result.
  • The problems arrive pre-framed. Every assignment tells you exactly what to build. But the actual job is figuring out what's worth building and why.
  • The one skill that now matters most — teaching yourself the next thing, with AI, faster than the field moves — is the one class can't grade.

So capable students graduate with no evidence they can take a vague, real-world problem, frame it, build a working AI system, and defend their decisions like an engineer. A certificate won't fix that. A real project, built under the critique of a working practitioner, will.

Not a class. A studio.

Shipyard AI is a three-week studio limited to five students per session, run by a working AI practitioner — not a training company.

Why not lectures? Because the lecture is the one part of education that AI has already made free. You can follow along perfectly and still freeze in front of a real problem. The only thing AI cannot replace is judgment: someone experienced looking at your work and telling you what's weak, what's promising, and what to do next.

Here's the deal instead:

An advisor, not an instructor

Your mentor works the way a Ph.D. advisor works: you own the project, they sharpen your questions, hold the bar at design reviews, and vouch for you when your work has earned it.

You learn; we critique

You teach yourself with AI using curated quest briefs that tell you what to master, not how. Then bring your work to the table for honest, specific critique.

You ship every day

Something small, something broken, something real — demoed to your cohort daily. Perfectionism dies in week one.

You review like working engineers

Structured peer review, written critiques, rotating partners. Giving sharp feedback is a job skill; you'll practice it daily.

What three weeks looks like

Week 1

Build fluency

Daily quest briefs take you from "I've used ChatGPT" to building with LLM APIs, retrieval, structured outputs, and evaluation — how to measure whether an AI system actually works.

Week 2

Find and frame a real problem

The differentiator. You'll source problems from the real world, vet them against a rigorous rubric (Is it valuable? Is the data accessible? Can you measure success? Is AI the right tool?), and decompose your chosen one into a buildable spec. Your spec gets a design review before you write a line of code.

Week 3

Build, evaluate, ship

Build your system, measure it, improve it based on what the evals tell you, deploy it, and present it at Demo Day to invited industry guests.

Sessions run in themed cycles — Foundations, Reliability & Evals, Agents & Automation — and campers can return for later sessions. Returning campers take on mentorship roles and ship at a higher bar each time.

What you leave with

A deployed project

Solving a real problem, with documented evaluation results — not a tutorial clone.

A three-week learning log

Showing exactly how you think and grow, day by day.

A demo-day presentation

Delivered to working engineers, defending your decisions under their critique.

Practiced job skills

Teaching yourself with AI, framing fuzzy problems, reviewing peers' work, defending decisions.

A cohort

Four peers who have seen your work up close, who you've reviewed, who will vouch for you.

Something to show

When you sit down with a hiring manager, you have evidence, not a certificate.

What we don't promise

❌ Internships or jobs

Nobody can honestly promise those — be suspicious of programs that do.

✓ A professional reference

If your work merits it, your mentor acts as a reference — someone from industry who can tell a hiring manager specifically what they watched you build and how you work. That reference is earned at the crit table, not purchased with tuition.

Who should apply

College students (any major — CS helps but isn't required) with:

Working Python basics

You can write a function with loops and dictionaries without help.

Self-driven mindset

You want critique, not hand-holding. You try things, get stuck, dig out, and show your work.

Generosity

You'll give honest, specific feedback to peers — at five students, you are each other's faculty.

Admissions

A short application task — use AI to teach yourself something you don't know, then write ~300 words on what you learned and what the AI got wrong — followed by a 20-minute conversation. We select for drive and curiosity, not credentials.

Logistics & tuition

Format

Live online (or hybrid)

3 weeks

20–30 hours/week

Cohort size

5

Maximum

Tuition (pilot)

$500

$100 returned on completion

Founding cohort pricing. Future sessions will cost more.

Rhythm

Sunday evening week-opening briefing → daily ship & demo → mid-week peer session → Friday demo & critique. Week 3 ends with Demo Day.

Mentor presence

Two live sessions per week (Sunday + Friday) + twice-weekly written feedback + bookable 1:1 office hours.

Tooling provided

Personal key to professional AI coding assistant + model access with session budget included. No subscriptions, no credit card, no setup costs.

For comparison: typical bootcamps charge $10,000–20,000 and spend it on marketing and placement teams. You're paying for one thing — a practitioner's attention on your work.

Your mentor — really, your advisor

Dr. Richard Xie

The role here is closer to a Ph.D. advisor or a senior engineer running design reviews than to a teacher at a whiteboard. You will not be taught at; you will be advised, challenged, and held to a professional bar.

Dr. Xie has spent 20 years leading AI and ML work in industry, has taught graduate AI courses at George Washington University, and has built and shipped multiple AI-powered applications in use today.

Shipyard AI started with one conversation: a talented CS junior who couldn't get an internship because she had no way to show what she could do. This program is the answer Richard wished he could have handed her.

Frequently asked questions

What does a typical day and week look like?

The studio runs on a weekly rhythm. Sunday evening we open the week with a live session (about an hour): I give a briefing on the week's focus, and everyone states what they'll demo by Friday.

Monday through Thursday you work on your own schedule — roughly 4–6 focused hours a day. Each day has a "quest": a one-page brief with a goal and artifact due that evening. You post a short async standup each morning, build during the day, and demo something every day — small, rough, and real all count. The point of daily shipping is to kill perfectionism early. You also keep a brief daily learning log, which I read and comment on.

Mid-week the cohort runs a peer session without me. Friday we close with a two-hour live session where everyone demos and gets critique — from me and from the other four students.

The three weeks have distinct shapes:

  • Week 1 builds your AI engineering fluency (working with models, retrieval, and especially evaluation — measuring whether an AI system actually works).
  • Week 2 you find and frame a real problem, with a real stakeholder, and get your plan through design review.
  • Week 3 you build, measure, and ship it — ending with Demo Day in front of invited industry guests.

What tools and technologies do I need?

You don't need to buy or subscribe to anything. The studio provides each student a personal key to our AI stack — a professional AI coding assistant (the same category of tool used in industry) plus direct API access to current models, with a usage budget included.

Around that you'll use the standard working-engineer toolkit: Python, VS Code, Git and GitHub, and a simple deployment platform for your capstone. If you already use tools like Copilot or Cursor, you're welcome to use them alongside.

Before Week 1 there's a short pre-work week where we get your full environment set up and verified, so no studio time is lost to installation problems.

How is progress evaluated?

There are no grades and no exams — this is a studio, not a class. Progress is measured the way it is on a real engineering team: by artifacts and critique.

Your work is reviewed continuously (daily demos, written peer reviews, my feedback on your log and designs), and two formal "gates" in Week 2 work like real design reviews — your problem choice and your build plan each need approval before you proceed.

Completing the program means delivering a concrete set of artifacts: a deployed capstone, an evaluation suite with measured results, your written problem analysis and spec, your learning log, your peer reviews, and your Demo Day presentation. Complete them all and $100 of your tuition is returned.

We don't promise jobs — what you leave with is a real, deployed project with measured results, and, if your work merits it, my professional reference.

What's the application process?

Three things:

1. A short note about you

Name, school, year, major, and a sentence or two on why you want to do this.

2. Commitment confirmation

Confirm you can commit 20–30 hours/week for the three weeks, and that you have working Python basics (you can write a function using loops and dictionaries without help — we'll send a short self-check).

3. The application task

Use AI to teach yourself something technical you don't currently know, then write ~300 words on what you learned and — importantly — what the AI got wrong or what you're still unsure about. There's no right answer; I'm looking at how you learn, not what you already know.

After that, we'll set up a 20-minute conversation. Five seats total, so I'd encourage you not to sit on it too long.

Apply now

Five seats. Next session begins Summer 2026.

Submit your application below or email shipyard.ai@outlook.com directly.

Keep it brief — 1–2 sentences
~300 words. This shows us how you learn, not what you already know.

We'll review your application and reach out within 5 business days. Five seats total — we encourage you not to sit on it too long.