English · 繁體中文
Enterprise IT operator building practical AI systems.
Multi-AI workflows · Agent governance · Durable memory
Visit Ai-Sister ↗ · Explore 9 case studies → · Email ↗
|
20 years Enterprise IT |
9 projects Public case studies |
4 AI systems Claude · GPT · Gemini · Grok |
LIVE / MULTI-AIAi-Sister ↗ One input. Four AI systems. |
A production web app that orchestrates ChatGPT, Claude, Gemini, and Grok through parallel chat, structured debate, an eight-step coding loop, roundtables, and focused single-agent modes.Hono React SQLite SSE Oracle Cloud ARMSee where it started → |
Nine public repositories across multi-AI experiences, agent infrastructure, security, and learning—55 GitHub stars at the July 12 snapshot. Star badges below stay live; each title opens a full case study with architecture, decisions, setup, current limits, and license status.
01 / MULTI-AI EXPERIENCES
Multi-AI Chat DesktopDESKTOP LOCAL-FIRST
|
The feature-complete v1.1.0 desktop edition: coordinate logged-in ChatGPT, Claude, Gemini, and Grok through five conversation-first workflows, then keep talking in the same local session—without model API keys.Tauri 2 React TypeScript RustSource ↗ |
Multi-AI ChatCHROME NO API KEYS
|
Turn the AI tabs already open in Chrome into one lightweight multi-model workflow surface.source v0.2.0Manifest V3 React TypeScriptSource ↗ |
AI BrainstormingWEB MVP ASYNC REVIEW
|
Submit an idea for a 5-, 12-, or 16-round review, leave, then return by private resume link or optional email report.source v0.1.0Hono React SQLite OpenRouterSource ↗ |
02 / AGENT INFRASTRUCTURE
openclaw-hermes-watcherGOVERNANCE OPERATIONS
|
Add a guarded, auditable evolution layer to an OpenClaw host while keeping upgrade authority with the human operator.Bash systemd OpenClaw HermesSource ↗ |
Clawd-LobsterCLAUDE CODE FILE-FIRST
|
A curated operating layer for reviewed specifications, persistent memory, reusable skills, workspaces, and multi-machine continuity.source v0.6.0 betaPython MCP SQLite Claude CodeSource ↗ |
Claude Idea Review SkillSKILL VALIDATION
|
Give product, creator, community, and business ideas an adversarial review before serious time is spent building them.installable skillClaude Code Skill Python Static reportsSource ↗ |
MCP Memory ServerMEMORY EARLY-STAGE
|
Model decisions, resolved issues, questions, and knowledge through an inspectable MCP memory server.source v0.1.0Python FastMCP SQLite OracleSource ↗ |
03 / SECURITY & LEARNING
Homograph Attack DemoSECURITY BILINGUAL
|
Make a visually deceptive Unicode domain understandable in a short, hands-on security-awareness lesson.HTML CSS JavaScript VercelLive ↗ · Source ↗ |
AI Beyblade X Field GuideKNOWLEDGE AI-ASSISTED
|
Turn scattered discussions into a score-aware 3v3 deck-building, buying, practice, and match-review framework.Markdown Research notes Traditional ChineseSource ↗ |
Browse the full project directory →
01 |
Operations first. Working for months in a real environment matters more than looking clever in a demo. |
02 |
Human authority. Agents can research and act within limits. Permissions and upgrade decisions stay explicit. |
03 |
Durable by default. State survives, work resumes, decisions stay inspectable, and failures leave evidence. |
Product systems
TypeScript · React · Hono · Tauri · Rust · Python · MCP · SQLite · Oracle · Bash
Enterprise operations
M365 · Entra ID · Intune · Google Workspace · Salesforce · Dynamics NAV · NIST CSF · ISO 27001
current-focus/
├── orchestrate Claude · GPT · Gemini · local models
├── govern defined roles · human authority · honest scoring
└── operate production systems · every day
The longer story.
Taipei, 1985–2008. Studied computer science at National Central University, ran campus networks, co-built a MUD that is still operating after 20+ years, and managed the university's highest-traffic personal BBS board.
United States, 2008–today. Earned a master's in computer information systems, then spent nearly two decades operating identity, endpoints, ERP, BI, CRM, and security programs inside real businesses.
Now. I apply that operator's perspective to AI products: not “what is the cleverest code?” but “what will people trust, adopt, recover, and keep running?”
TAIPEI → UNITED STATES
ted-h.com · ted@ted-h.com


