Projects

Three products, all three running today — two inside a company, one built for myself.

Self Checkout

Zero-hardware QR checkout · KIRAPAY · 2026.04 — in daily use

Visit
A customer's phone: HKD $128.50 to KIRAPAY, pick a wallet

The call I made

Refused a hardware terminal. A static QR sticker is worse on paper — no screen, no card reader — but it turns deployment from an install visit into one sticker, and staff training into one sentence.

My part
Solo: flow design, build, and in-person merchant onboarding
Scale
Physical Hong Kong shops, daily, since 2026.04
What came of it
Still in daily use 4 months after launch, with no redeployment and no retraining.
  1. Idle: the shop screen waits

    Idle: the shop screen waits

  2. Staff key in the amount

    Staff key in the amount

  3. Customer pays from their own phone

    Customer pays from their own phone

Sippa

Creator tipping, 0→1 · KIRAPAY · Launched 2026.08

Visit
Sending a creator a $10 matcha, with a note attached

The call I made

Cut the MVP down to the money path — email notification, a mobile step-flow, and the UX fixes that were blocking payment. Everything else went to v1, which proved out the payment integration weeks before the product was pretty.

My part
Built it alone, then led the team of three that took it on
Scale
Zero to launch; crypto rails wrapped for creators who have never held a wallet
What came of it
Launched 2026.08. The cross-team payment integration landed without me having any authority over that team.
  1. The pitch, in one screen

    The pitch, in one screen

  2. A creator's page

    A creator's page

  3. Tipping, step one

    Tipping, step one

  4. Ready to send

    Ready to send

Personal multi-agent system

Self-built · plans, reviews and executes projects autonomously · 2026 — present

The agent stops: one decision needed, three options, each with its downside spelt out
My own tool, so the interface is in Chinese — and the project names in these shots are stand-ins, swapped in before the screenshots and swapped back after.

The call I made

Chose guardrails over reach. Every irreversible action stops for a human and every repository takes one writer at a time, which makes it slower than it could be — and is exactly why it can be left running overnight.

My part
Built it myself, for myself — then put it to work on a production product
Scale
Runs round the clock on my own projects and on Sippa in production
What came of it
86 bugs closed overnight, five review-ready posts a day, partner outreach drafted on schedule — work that would otherwise need a QA engineer, a content assistant and a BD researcher.
  1. Every project, this morning's brief

    Every project, this morning's brief

  2. What ran overnight

    What ran overnight

  3. A plan, before it is allowed to run

    A plan, before it is allowed to run

Skills

Product
  • End-to-end lifecycle (0→1 to launch)
  • roadmapping
  • PRD & requirements
  • success metrics & data-driven decisions
  • MVP scoping
  • cross-functional leadership
  • go-to-market
Build
  • TypeScript
  • React / Next.js
  • Supabase
  • Vercel
  • REST API integration
  • Git & CI
  • Python
  • SQL
AI & Automation
  • Agent orchestration & guardrail design
  • LLM-assisted development (Claude)
  • workflow automation (n8n)
  • AI batch pipelines
  • prompt engineering
Design
  • UX/UI & interaction design (Figma)
  • design systems
  • brand identity & visual design
  • UI copywriting
  • data visualisation
Domain
  • Web3 payments
  • crypto checkout & QR payment flows
  • fiat on/off-ramp partnerships
  • creator economy
Languages
  • Mandarin (native)
  • English (working proficiency)
  • Cantonese (basic)