AI Product Manager · Building AI-native products and original interactive experiences
LLM · Agent · Memory · Evaluation · Product Prototyping · Creative Coding
I turn real-world problems and playful ideas into working products. My work spans AI workflows, context and memory design, evaluation, privacy-aware experiences, and original interactive prototypes.
AI-powered work memory and reflection system
WorkLog turns fragmented daily notes into structured work records, project progress, risks, reviews, reports, and career evidence.
What it does
- Structures daily work, todos, risks, notes, life records, goals, ideas, and reports.
- Generates AI reviews for daily records, weekly progress, projects, and reports.
- Aggregates records into project dashboards with progress, risks, follow-ups, and archived states.
- Supports weekly reports, project reviews, stage summaries, and custom reporting periods.
- Keeps model API keys local while allowing optional Cloudflare D1 sync for user records.
AI / Product Concepts
AI review · work memory · structured records · report generation · multi-model support · local-first UX · privacy boundary
Stack
React Native Web / Expo, JavaScript, Cloudflare Pages Functions, D1, PWA
An original hand-painted garden defense game for the web
Live battle and the hand-painted garden — click either image to play in the browser.
Garden Wardens is a five-lane tower-defense experiment built around a living garden, day-and-night strategy, changing weather, plant fusion, automation tools, and animated enemies.
What it explores
- Combines hand-painted art direction with responsive WebGL gameplay.
- Uses plant fusion, energy automation, mystery rewards, and global upgrades to create strategy beyond straightforward placement.
- Treats motion, combat feedback, onboarding, difficulty, and desktop usability as product-design problems—not just implementation details.
- Ships as an open-source browser game with a playable GitHub Pages build.
Product / Game Concepts
game UX · systems design · progressive onboarding · feedback & animation · difficulty balancing · open-source iteration
Stack
JavaScript, Phaser, WebGL, Vite, GitHub Pages
A cross-platform desktop break reminder for screen-heavy work
Tonicue runs quietly in the background, distinguishes active work from idle, lock, and sleep states, and delivers small always-on-top reminders without interrupting the user's flow.
What it demonstrates
- Windows tray and macOS menu-bar background behavior.
- Active screen-time tracking with idle, lock, and sleep handling.
- Guided rest flows, snooze and focus modes, local statistics, and local-only settings.
- Installable preview builds for Windows and macOS Apple Silicon.
Stack
Electron, JavaScript, HTML, CSS, electron-builder
Personal learning product for focused English reading
English Reader is a responsive web reader for reading English books page by page, looking up words inline, and collecting vocabulary cards.
What it does
- Provides page-by-page reading across mobile and desktop.
- Lets readers tap words for inline lookup without breaking the reading flow.
- Collects vocabulary cards with examples for later review.
- Supports browser-local PDF, TXT, and Markdown import.
- Keeps the public frontend separate from private account, sync, upload, and dictionary services.
Product Concepts
learning workflow · reader UX · vocabulary capture · local file import · public/private boundary
Stack
TypeScript, React, Vite, pdf.js, Cloudflare Pages
Mobile-first digital wardrobe and outfit tracking app
My Wardrobe is a personal consumer product for organizing clothes, creating outfits, recording wear history, and understanding wardrobe usage.
What it does
- Manages clothes with photos, categories, seasons, colors, tags, brands, sizes, price, and notes.
- Creates outfit combinations and connects outfits back to individual clothing items.
- Records daily wear history and shows usage, recent wear, cost per wear, and wardrobe insights.
- Stores data locally by default with JSON, CSV, and image export paths.
- Includes an optional Cloudflare D1/R2/Pages Functions architecture for future cloud sync.
Product Concepts
consumer product thinking · mobile-first UX · local-first storage · image/data management · usage insights
Stack
TypeScript, React, Vite, IndexedDB, localStorage, Cloudflare architecture
| Area | What I'm exploring |
|---|---|
| Agents | Tool use, workflows, agent UX, and human handoff points |
| Memory | Short-term and long-term memory, retrieval, user control, and deletion/correction flows |
| Evaluation | Prompt/model comparison, regression checks, quality criteria, and product reliability |
| Context | Context construction, information prioritization, and task-specific compression |
| FinTech AI | AI-native financial product experiences with compliance, risk, and trust boundaries |
| Product Systems | From user problem discovery to prototype, architecture, launch, and iteration loops |
- Building: WorkLog AI capability upgrades around memory, review quality, and evidence generation.
- Shipping: Garden Wardens gameplay, animation, difficulty, and visual-feedback improvements through rapid playtest-driven iteration.
- Exploring: AI Prompt / Skill / Eval Studio for comparing prompts, models, and workflow outputs.
- Planned: Agent Memory Lab for experimenting with user-visible memory, retrieval, and correction.
- LLM fundamentals and Transformer/GPT internals
- Agent systems, tool use, and context engineering
- LLM evaluation and reliability practices
- PyTorch fundamentals for understanding model behavior more deeply
Product & AI
LLM APIs, prompt engineering, AI evaluation, agent workflows, memory UX, privacy boundaries
Build
TypeScript, React, JavaScript, Python, Node.js, Vite, Expo Web
Infra
Cloudflare Pages, Pages Functions, D1, R2 architecture, IndexedDB, localStorage, local-first product design
| Category | Repositories | Portfolio role |
|---|---|---|
| Flagship Products | worklog | Main AI product portfolio project |
| Product Experiments | garden-wardens, screen-break-reminder, english-reader, my-wardrobe | Original products, games, and prototypes, ordered by portfolio signal |
| Forks / References | dify, happy-llm, self-llm, llm_interview_note, daily_stock_analysis, freeCodeCamp, LlamaIndex-Tutorials, newspaper |
References and study materials, not presented as original products |
I care about products that make AI useful in real workflows: clear user problems, explicit data boundaries, working prototypes, measurable quality, and interfaces that help people understand and control AI behavior.