Purdue BME · gap year through spring 2027

I run a one-person company staffed by AI agents.

I'm Kevin Yu. I take products from spec to production on my own, with Claude Code and a crew of AI agents doing the work around me. In mid-2024 I had never written code. Today I'm building a SaaS, running a book publishing operation on Amazon, making AI video and shipping healthcare demos at hackathons.

Human decisions Agent work

Company OS Running
How Kevin's one-person company is wired Kevin directs eleven AI agents and tools, including Claude Code, Codex and a Grok Bot crew. They work across ten lanes such as Hirvo, Bookflow and Amazon KDP, YouAreExtinct and UHOME.
Claude Code Deep repo work: the Hirvo rewrite, Bookflow hardening, sites and 3D builds.
  • 2 yrsfrom no coding background to shipping production apps
  • 9coloring books live on Amazon
  • 30/30books made in a 30-day sprint
  • 10lanes running in parallel, one operator

Selected work

What I've shipped, with the how left in.

Five projects across SaaS, publishing, video and healthcare. Each one shows the system behind it.

In production buildCo-founder & CTO

Hirvo

Job search for international students that knows which employers sponsor visas, and rewrites resumes without inventing anything.

  • Built solo in Next.js, TypeScript, Supabase and the Anthropic API.
  • Anti-fabrication resume engine with a three-layer ATS risk model.
  • Five PRD versions and an engineering handoff pack before the build.
  • Set up the GitHub org and branch protection, then onboarded engineers.

Repo available on request

hirvo · resume rewrite

Data Engineering Intern

Seattle, WA · Summer 2027

Sponsors visasOPT eligible
Source

Built a Flask app to track lab inventory.

Draft

Built and deployed a Flask app on AWS serving 10k users.

Check

✕ “AWS” not in source✕ “10k users” not in source

Final

Built a Flask web app that tracks inventory for a research lab.

Illustration of the anti-fabrication check

Live on AmazonPublisher & toolmaker

Bookflow + Amazon KDP

A small publishing house behind the Snuggle Crew coloring books: nine titles on Amazon, a team to run it, and my own tools for making them.

  • Built an LLM coloring-book generator, then Bookflow to go from create to metadata.
  • Ran a 30-books-in-30-days sprint.
  • Run a multi-person publishing team.
  • Agents handle keyword research and read Amazon Ads campaigns. I approve every publish and every dollar of ad spend.

9 titles liveHover a spine

  1. Cindy niche
  2. Tim keywords
  3. Bookflow create
  4. Jennifer art
  5. Kevin publishes
  6. Alex ads read
Live on YouTubeMade solo with AI

YouAreExtinct

Cinematic AI stories about extinct animals. The channel is live, and it doubles as my portfolio for brand video work.

  • Each film starts from a reference, then Claude writes a shot-by-shot playbook.
  • The same pipeline makes AI ads for brands, without a production crew.
  1. Reference pull
  2. Claude shot list
  3. AI video
  4. Edit cut
  5. Kevin publishes
1 · Snap
必理痛 PanadolParacetamol 500 mg
2 · Match

Same active ingredient: paracetamol

3 · Speak
粵語

Illustrative example

HackathonAIx Origin · Vital

對藥

A medication check built solo at Cyberport. It photographs discharge sheets and medicine boxes, matches drugs by active ingredient with a deterministic rules engine, and reads the result aloud in Cantonese.

Model → render loop

DeliveredWeb + 3D

UHOME

Student apartments near the University of Alberta in Edmonton. I built the real estate website and did the 3D modeling and renders for the tours.

How it runs

Five steps, whatever the lane.

  1. 01

    Try to kill the idea

    Every idea faces one test: could a general chatbot already do this? If yes, it dies that day. Whatever survives gets built properly.

    Could a general chatbot already do this?

    Idea

    A PDF summarizer app

    Killed
    Idea

    Job matching that knows which employers sponsor visas

    Build
  2. 02

    Write the spec

    I write PRDs for coding agents, with typed interfaces and Given/When/Then acceptance criteria. Hirvo went through five versions and an engineering handoff pack before the build started.

    prd.mdexample
    interface RewriteRequest {  resume: Resume  target: JobPosting} Scenario: rewrite never invents experience  Given a resume with no AWS experience  When  it is tailored to a cloud role  Then  no bullet mentions AWS  And   every claim traces to the source
  3. 03

    Hand it to the crew

    Claude Code and Codex do the deep repo work. A Grok Bot crew runs research, art briefs, ads reads and ops, with Notion scoreboards and handoffs so nothing lives only in my head.

    Kevinassigns the work
    • Claude CodeRepo workdone
    • CodexRepo workdone
    • CindyNiches, keep or skipdone
    • TimReverse-ASIN, longtailsdone
    • JenniferArt packs, A+ pagesdone
    • AlexAds read: hold or changedone
    • BobBookflow toolingdone
    • SadieStory-IP directiondone
  4. 04

    Keep the gates human

    Agents research, draft and recommend. Anything that publishes, spends money or needs a 2FA code waits for me.

    Publish to Amazon KDPproposed by Bob
    Approved by Kevin
    Raise ad budgetrecommended by Alex
    Approved by Kevin
    2FA sign-inrequested by Tim
    Approved by Kevin

    Agents can't publish or spend on their own.

  5. 05

    Ship, then distribute

    AI made building cheap. Getting it in front of people is still the scarce skill, so every lane ships with a channel attached.

    ✓ build passed

    ✓ deployed

    → distribution

    • Amazon Adsvia KDP Marketing
    • YouTube@YouAreExtinct
    • Xbuild in public
    • WeChat 小程序拼豆 store
    • LinkedInroles and clients

AI made making cheap. Getting things in front of people is still the hard part.

The whole board

Ten lanes, one operator.

Everything I'm running right now: who it's for, how it runs and where it stands. Open a row for details.

  • Who it's for
    International STEM students aiming at US roles, who need sponsorship clarity and honest, strong applications instead of spam-apply bots.
    How it runs
    I'm co-founder and CTO. Built with Claude Code, Codex and the agent crew. The loop: rewrite system, eval sets, co-founder feedback, retune, pricing, sell.
    Where it stands
    In production build with no public URL yet. The repo is available on request.
  • Who it's for
    Buyers of cozy, seasonal and wholesome coloring books. On the seller side, the pain is slow creation, keyword research, ads ops and art throughput.
    How it runs
    Cindy and Tim run niche and keyword research, Jennifer makes art packs, Bob builds Bookflow tooling and Alex reads ad campaigns. A Notion hub is the source of truth. I hold the gates: publish, 2FA and ad spend.
    Where it stands
    Nine books live, most recently Cozy Thanksgiving in October 2026. Now shifting from volume to quality and wholesome story IP. bookflowai.site
  • Who it's for
    Brands that want short AI-shot ads without hiring a full crew.
    How it runs
    Download a reference, have Claude write a shot-by-shot playbook, edit, then deliver to Drive. YouAreExtinct is the portfolio.
    Where it stands
    First paying lead in progress, with a coffee brand demo next.
  • Who it's for
    Viewers who like immersive “last of their kind” stories. It's also my AI video portfolio.
    How it runs
    Run solo on the AI video stack, from shot list to final cut.
    Where it stands
    Live. youtube.com/@YouAreExtinct
  • Who it's for
    Students and focus workers who want lyric-free background music for long sessions.
    How it runs
    Run solo. The music is made with an AI-assisted production workflow.
    Where it stands
    Music is ready. The first upload is next.
  • Who it's for
    People who already love 拼豆 (fuse beads) and need a store they can find and an easy way to visit.
    How it runs
    A team-owned store. I own growth and tech: social, the 小程序 and the websites, built with AI-assisted workflows.
    Where it stands
    Just started.
  • Who it's for
    Consistent on-camera talent without full influencer ops. It's also how my cofounders and I learn distribution by doing.
    How it runs
    Built with two cofounders. Soul 2.0 looks are trained on Higgsfield, and agents coach looks, cuts, stories and art.
    Where it stands
    Training is done and production is ready to start.
  • Who it's for
    Students and families looking for clear listings and good visual tours near the University of Alberta.
    How it runs
    I built the real estate website and did the 3D modeling and rendering, using AI where it helped.
    Where it stands
    Delivered. View the demo
  • Who it's for
    People who want real build logs from Hirvo, Bookflow, AI video and the rest. For me, it's distribution practice.
    How it runs
    The plan came out of a playbook I studied plus the Big Three agents. Grok Bot drafts posts, and I hit publish.
    Where it stands
    Plan locked. The handle goes live soon.
  • AIx Origin · Vital
    對藥, a medication check for patients and carers: same drug with two names, and leftover boxes turned into a card for the pharmacist.
    复客松 · 华润三九杯
    An AI healthcare build for elder medication confirmation. Finished in September 2026.
    Fudan
    Received a recognition certificate. I entered team events solo and still shipped working demos.

Tools and range

What I build with, and what else I bring.

Build and ship

  • Next.js
  • TypeScript
  • Supabase
  • Expo
  • VPS deploys
  • GitHub orgs

Production repos with branch protection and engineer onboarding.

AI in the loop

  • Claude Code
  • Codex
  • Cursor
  • Custom Claude skills
  • Anthropic API
  • Grok Bot agents

Agents work inside every lane, with Notion scoreboards and handoffs.

Media and compute

  • ComfyUI
  • DGX Spark
  • Tailscale
  • MiniMax TTS
  • Higgsfield Soul 2.0

ComfyUI runs on a DGX Spark that I reach over Tailscale.

Founder operations

  • Cap tables
  • IP ownership
  • SAFEs
  • 83(b) elections

Handled myself, alongside the product work.

Experience

  • Co-founder & CTO, Hirvo
  • Project supervisor
  • Startup advisor

Supervised a project for a US company and advised an MCN startup.

Range

  • Biomedical engineering
  • Statistics
  • Healthcare AI
  • English
  • Mandarin

Working across Canada, the US, China and Hong Kong.

Contact

Hiring for healthcare AI, product or AI engineering? Need an AI-shot ad? Let's talk.

Message me on LinkedIn

Hirvo's repo is available on request. I work in English and Mandarin.