Intent → contract
Natural language becomes an editable, bounded structure — never an opaque command.
Senior iOS Engineer · Taiwan
I build products where intelligence is infrastructure.
10+ years shipping production iOS products. Now using AI-assisted development to build products with clear validation, user control, and recovery.
- 10+
- years shipping
- ~100K
- DAU · real-time
- ~20K
- DAU · wallets
01Intent
02Structure
03Evidence
04Recovery
AI ASSISTED
System
AI, STRUCTURALLY
AI is not a feature. It is a new product contract.
The interface can feel fluid. Underneath, authority, truth, recovery, and evidence stay explicit.
AI proposes. Product owns.
Validation, rendering, persistence, and recovery remain deterministic product responsibilities.
Human authority
The user can inspect state, understand the next action, and stop unsafe mutations.
Evidence over theater
Receipts, checks, and recovery paths matter more than an agent merely appearing busy.
SELECTED SYSTEMS
Three expressions of one idea.
Moment OS
From a life intent to one truthful next action.
Moment is a React Native, Expo, and TypeScript product that turns natural-language intent into understandable, editable preparation. AI suggestions stay separate from saved data so user-approved plans remain recoverable and confirmed results stay distinct from unverified suggestions.
ROLE
Creator · Product, architecture, React Native
ProductDev
A workbench for automation you can actually trust.
ProductDev observes and safely manages repository automation without taking control from the existing scheduler. GitHub Issues remain the work contract; changes are guarded, previews match the real action, and every supported action leaves a permanent record.
ROLE
Creator · Product, systems, macOS
Continuity. Truth. Pace.
The tooling behind long-running AI work.
A family of open tools turns fragile agent sessions into reviewable engineering: Context Handoff preserves acceptance criteria across fresh threads, Task ETA Tracker forecasts from observable milestones, and Moment Monitor exposes live state without gaining mutation authority.
SELECTED COMPANY PRODUCTS
Products I contributed to in production.
Selected contributions made as part of engineering teams across self-custodial wallets, real-time betting, and connected mobility.
01 / WALLET
Crypto.com Onchain
Owned iOS Earn across multiple blockchain protocols. Shipped staking, rewards, position management, and Buy & Swap for a self-custodial wallet.
- ~20K
- DAU
- iOS
- Ownership
- Multi-chain
- Protocols
02 / REAL-TIME
SportyBet
Built live odds, open bets, and betting-history synchronization. Prevented duplicated WebSocket updates across multiple betting slips and evolved legacy modules.
- ~100K
- DAU
- WebSocket
- Live state
- RxSwift
- Migration
03 / MOBILITY
KINTO
Delivered vehicle usage tracking, vehicle status, and in-app guidance for a connected vehicle application on a focused three-month production cycle.
- 3 mo
- Contract
- Vehicle
- Live state
- iOS
- Production
HOW I BUILD
Product clarity before model cleverness.
- 01
Define the truthful outcome
Name the user problem, observable done condition, and what the product must never claim.
- 02
Separate intelligence from authority
Let models interpret and propose; keep validation, persistence, and mutations behind explicit contracts.
- 03
Design failure and recovery
Treat checkpoints, stale writes, degraded modes, and receipts as product surfaces — not backend trivia.
- 04
Verify the experience
Prototype the interaction, test deterministic gates, then measure whether the system helps the user move.
BUILD THE NEXT SYSTEM