I've shipped car recognition at 100+ production sites.

End-to-end: custom model training, camera hardware selection, edge deployment, ops dashboards, and continuous optimisation — across Hong Kong, Japan, Australia, and the US. I've handled the failure modes that kill LPR projects: glare, rain, angle, plate-format variety, and 24/7 uptime expectations.

100+

Production sites running my recognition system

4 markets

HK · Japan · Australia · US — different plate formats & conditions

End-to-end

Model → hardware → deployment → ops platform

01 — Smart Parking

Computer Vision · Deployed at Scale

Smart Parking

Fully automated operating system for smart parking across HK, Japan, Australia, US. 100+ sites.

Led end-to-end development: model training, camera hardware.
Operation: ops dashboard, site deployment, optimisation.

Scope

Custom vision model 100+ Sites HK · Japan · AU · US Camera Hardware Edge Deployment

Car recognition program

Web portal & custom dashboard

Custom space camera deployment - Japan

02 — PitchSight

Football · Computer Vision

PitchSight

Turn football match videos into tactical evidence for training

Custom-trained player/ball trackers, team classification, ball action spotting, shot detection
Dataset collection, annotation, and evaluation
Portable camera for auto-capture → custom-built hardware setup + broadcast-ready video with auto ball following

Proof

Match footage → clips & reports Tactical map Custom event detection Real-world pitches

Camera auto-film and follow

Tactical map with position tracking

PitchSight — custom event tracking including ball and players

Custom event tracking inc. ball, players

03 — AirDrum

Computer Vision · Real-time

AirDrum

Custom-trained vision model + Python audio engine. Built for drum learners who want to practice at home.

Detects drumstick position and motion from a webcam feed. Triggers drum samples in real time.
Built end-to-end: dataset collection, model training, detection logic, audio output.

Technical

Custom Model Object Tracking Python

Demo clip

Custom model training

Testing

04 — Personal Agentic System

AI Agents · Automation

Personal Agentic System

A personal mission control infrastructure that orchestrates AI agents across research, coding and planning.

Cursor, OpenClaw, Telegram — coordinated into daily autonomous workflows.
Daily autonomous tasks: topic tracking, competitor monitoring, client follow-up etc.

Technical

Agent Orchestration Cursor / OpenClaw Telegram n8n

Mission control dashboard for agent management

Bookmark tool as LLM wiki

Automated job posting monitoring

Why I fit this role

What I bring to your team.

Production-ready

I led car recognition from new product development through sales, deployment, and optimisation — I know what production-grade means for parking operators.

Hardware-aware

Camera selection, mounting geometry, edge compute — I've made these calls at scale, so you don't pay for trial-and-error.

I stay for the optimisation

Accuracy tuning per site, exception handling, monitoring dashboards — the part that decides whether the system actually gets renewed.

How I work

LPR & mobility

Custom plate recognition, camera hardware, edge deployment, 100+ live sites

Computer vision

Detection, tracking, model training, per-site tuning, ops dashboards

Delivery

End-to-end — dataset, model, deployment, dashboard, optimisation

About

Operator turned product builder. I led smart-parking vision systems from model training through deployment and optimisation across four markets.

Comfortable across product, ops, and engineering — five years as product lead before going independent.

Where I am: Hong Kong. Remote, timezone-flexible.

Background

Founder, Vectory Labs — Custom visual intelligence for the physical world.
Product & Solutions Lead, UrbanChain (HK) — Smart parking across HK, Japan, AU, US.
Manager, Klook (HK) — Global Travel platform, 10+ markets.
Head of Driver Operations, Lalamove (HK) — Ride-hailing, 10,000+ driver community.
Senior Consultant, EY (HK) — Risk Advisory.

Get in touch