AI & ML
In productionBuilt LungLens — a full-stack AI health literacy tool using Hugging Face inference API (live at lunglenshk.vercel.app). Helped design an LLM-based monitoring agent at AAHK using Power Automate + Dynatrace API + ServiceNow.
Selected Work
Selected work
Problem → what was built → outcome
Career and product work across AI, mobile, and enterprise systems. Each card is grounded in a real role, stack, and result — not self-rated scores.
Focus areas
Where I usually contribute.
Product Management
14 years deliveringFourteen years across Apple, HKJC, Cathay Pacific, and AAHK. At Cathay Pacific, the mobile app grew from ~10,000 to ~120,000 monthly active users (12x) during my ownership.
Event-Driven Architecture
Deployed at HKJCAt HKJC, worked on a Solace migration — moving infrastructure from traditional REST APIs toward event-based messaging.
Edge Computing
Shipped at AppleAt Apple, contributed to docserverless — moving numerical computation from server-side to edge user devices.
Opening line
Here’s a selection of work I’ve been part of — some enterprise delivery, some products I’ve built myself.
AI & machine learning
Production agents and full-stack ML products
LungLens
In production
Full-stack AI health literacy tool — live in production
- Role
- Solo — design, frontend, backend, deployment
- Stack
- React + Python + Hugging Face Inference API, deployed on Vercel
- Problem
- Patients struggle to understand chest X-ray results in plain language.
- What I built
- Bilingual (EN/繁中/簡中) educational tool that helps patients understand their chest X-ray results. Users upload X-ray images for anatomy-focused, plain-language educational analysis — explicitly not a diagnostic tool.
- Outcome
- Live at lunglenshk.vercel.app
MLLLMProductionReact + Python
End-to-end AI/ML app I designed and shipped alone — Hugging Face inference, Vercel frontend, bilingual UX.
AAHK LLM Monitoring Agent
In production
Automated infrastructure health reporting for senior management
- Role
- Product lead — gathered requirements, designed approach, managed vendor delivery
- Stack
- Power Automate + Dynatrace API + ServiceNow integration
- Problem
- Senior management relied on manual infrastructure health reports (CPU, RAM, downtime).
- What I built
- LLM-based monitoring agent that automates infrastructure health reporting via Power Automate, Dynatrace API, and ServiceNow. Led product design and vendor delivery — not coded solo.
- Outcome
- Estimated 80% reduction in manual reporting time
LLM AgentsEnterpriseAAHK
Enterprise AI delivery at AAHK — requirements, design, and vendor management, not a solo coding story.
Product & apps
Mobile and consumer products — PoC to live
Cathay Pacific Mobile App
In production
Airline mobile app — lifecycle ownership over six years
- Role
- Product Manager / Product Owner
- Stack
- Mobile app integrated with legacy Passenger Service Systems (PSS)
- Problem
- Cathay needed a modern mobile experience wired into legacy PSS backends.
- What I built
- Led product ownership for the Cathay Pacific mobile app — backlog, stakeholder negotiation, and XML/JSON integration with legacy PSS.
- Outcome
- Grew monthly active users from ~10,000 to ~120,000 (12x growth)
ProductMobileCathay
Six years of lifecycle ownership — not a launch-and-leave story. MAU grew roughly 12x.
北辰 (North Star)
In Progress
Bazi/ZiWei astrology app inspired by Co-Star — targeting iOS
- Role
- Solo founder — design, architecture, development
- Stack
- React Native (planned)
- Problem
- Chinese astrology (Bazi/ZiWei) lacks a modern, Co-Star-quality mobile experience.
- What I built
- Designing North Star — user stories mapped in Notion, preparing the React Native codebase for iOS.
- Outcome
- Design phase — in progress
In ProgressMobileiOS
Current build focus — astrology product design for iOS, not vaporware: stories mapped, codebase next.
Systems & architecture
Event-driven, edge, and enterprise delivery
HKJC Solace Migration
In production
Event-driven messaging for regulated real-time systems
- Role
- Product Manager / IT Solutions Analyst
- Stack
- Solace Event Mesh (Pub/Sub)
- Problem
- Core infrastructure depended on traditional REST API patterns under high concurrency.
- What I built
- Led Solace migration — moved core infrastructure from traditional API architecture to Solace event-based messaging for real-time odds and betting flows.
- Outcome
- Event-based messaging in a regulated, high-concurrency environment
Event-DrivenSolaceHKJC
HKJC Solace work — REST to event mesh in a compliance-heavy setting.
Apple docserverless
In production
Edge computing for numerical workloads
- Role
- Product Manager (Engineering Project Manager)
- Stack
- Edge / serverless — computation on user devices
- Problem
- Numerical computation ran server-side, adding latency and infrastructure cost.
- What I built
- Led the docserverless project — moved numerical computation from server-side to edge user devices.
- Outcome
- Computation shifted to the edge on user devices
Edge ComputingAppleServerless
Apple edge project — architectural review discipline applied to moving compute off the server.
Also worth a look
Side projects and experiments
Hermes
Side Project★ 3
ExpiryScanner — barcode + AI mobile app
- Role
- Solo builder
- Stack
- React Native + Expo + GPT-4o-mini + Supabase
- What I built
- Barcode scanning, AI product ID, Supabase backend, and push notifications — full product loop.
App DevGenAIMobileReact Native + Expo
AI Agent X-Ray
Side Project
Interactive agent tool-calling visualizer
- Role
- Solo builder
- What I built
- Side-by-side chat and x-ray view of LLM tool calling — how I explain agent architecture to stakeholders.
LLM AgentsAIArchitectureHTML + JS
Product Tax Deduction Log
In Progress
Mobile app for product purchases and tax deductions
- Role
- Solo builder
- Stack
- React Native + Expo
- What I built
- Personal finance utility for tracking product purchases and tax-deductible expenses. App Store submission in progress.
In ProgressMobileFinanceReact Native + Expo
localLLM
Side Project★ 2
Local LLM experimentation
- Role
- Solo builder
- What I built
- Hands-on local LLM evaluation — latency, cost, and fit before vendor recommendations.
MLLLMEvaluationPython
Closing line
Happy to go deeper on any of these — architecture decisions, stakeholder trade-offs, or how I’d approach your problem.
Short on time? Show LungLens and the AAHK LLM agent, then link to full experience.