— Sean
From open-source gateways to macOS apps, from protocol layers to intelligence systemsFrom open-source gateways to macOS apps, from protocol layers to intelligence systems
Over the past two years, I've built 6 AI products and 3 open-source repos on my own. Not demos — things people actually use.Over the past two years, I've built 6 AI products and 3 open-source repos on my own. Not demos — things people actually use.
Production-grade tools in public repos.Real systems in public repos. Not demos — production-grade tools you can clone, deploy, and use today.
Open Source · Luxury Hotel AI Agent Protocol
Turns luxury hotel search, comparison, and quoting into agent-ready MCP tools.An MCP server that brings AI agents into luxury hotel workflows. Not a chat wrapper — it decomposes search, benefit comparison, quoting, and checkout handoff into agent-callable tools with Streamable HTTP support.
Open Source · Real-Time Hotel Rate Comparison
Real-time price comparison across 4 OTAs with member pricing support.An MCP server that enables AI agents to compare hotel prices across Trip.com, Booking.com, Agoda, and Google Hotels in real time. Supports demo and live modes with Playwright browser automation, session persistence for member pricing, and batch queries for up to 10 hotels concurrently.
Open Source · API Protocol Translator
Lets Claude Code talk to any OpenAI-compatible gateway via real-time protocol translation.Corporate Claude endpoints speak OpenAI chat/completions, but Claude Code only speaks Anthropic Messages. This reverse proxy translates both directions in real time — requests, responses, tool calls, streaming SSE, and token counting. Zero client patches, zero upstream changes.
Not screenshots. Real AI systems already in motion.These are not portfolio screenshots. They are working systems already touching the real world. Some prove protocol design, some prove judgment generation, and some prove business semantics plus hybrid control.
Turn protocols into interfaces.Turn MCP, APIs, and agent routes into a business interface.
Put AI inside the loop.Put AI inside a real operating loop instead of a chat wrapper.
Turn judgment into systems.Convert signals, content, and semantics into executable systems.
I compress models, workflows, data, and human action into interfaces that can enter real business systems.
Lots of people can call a model. Very few can turn it into something that actually works.
My approach is simple: figure out what problem the product solves, how engineering makes it real, and where the system should stop.
From SiftGate's open-source gateway to Bitvoya's hotel protocol layer to Morrow's macOS app — each one was built from scratch by my own hands, not drawn on slides.
What I do is turn complex technology into products that business understands, engineering can build, and users can trust.
Define what AI should not own before deciding what it should do.
Orchestrate models, workflows, data, and human action into one loop.
A product earns trust when it knows exactly where AI should stop.
Not slogans. The real rules behind how I build products and AI workflows.These are not slogans. They are the operating system behind how I design products, protocols, and AI workflows.
Workflow first, model second.Define the workflow, roles, and responsibility lines first. The model comes after. Without workflow, AI is still theater.
Structure before magic.If it can be structured, it should not stay fuzzy. Clear semantics create stable systems and faster collaboration.
Trust needs a boundary.AI can generate, suggest, and compare, but it should not cross the trust line. A product feels mature when it knows where to stop.
Value lives in real use.Real value is not in the benchmark. It is whether the system enters the business, produces outcomes, and keeps being used.
I care about four things: interface, workflow, semantics, and boundary.I do not think in skill trees. I care about four things: whether the interface is clear, the workflow is closed-loop, the semantics are reliable, and the boundary deserves trust.
Speed gets you to the demo. Judgment and boundary design make it worth shipping.Speed gets you to the demo. Judgment, structure, and boundary design are what make the system worth shipping.
I turn vague business intent into tools and states an agent can execute.I compress vague business intent into tools, states, and actions an agent can actually execute, instead of trapping AI inside a chat demo.
This is how MCP and business APIs start speaking the same language.This is how MCP, internal operators, and business APIs start speaking the same language.
I place the model at the right step so code, rules, and humans can work with AI.I place the model at the right step and let code, rules, human action, and AI cooperate, rather than asking the model to carry the whole system alone.
I build durable workflows, not one-shot prompt magic.I build durable workflows, not one-shot prompt magic.
I make pricing, perks, and edge conditions explicit so agents can work inside real business context.I structure pricing, perks, inventory, language, permissions, and edge conditions so the agent can operate inside real business context.
Especially useful when AI touches commerce, content, or operational complexity.Especially useful when AI touches commerce, content, or operational complexity.
I let AI generate and prepare, while keeping high-risk submission inside trusted surfaces.I let AI generate, suggest, prepare, and compare, while keeping high-risk submission inside trusted surfaces and deterministic systems.
A mature product knows what should never be automated.A mature product often feels mature because it knows what should never be automated.
AI is not the first complex system I have had to make work.
Before AI, I spent years shipping results inside travel, logistics, and other high-constraint business systems.Before AI, I spent years inside travel, logistics, supply, commerce systems, and cross-functional execution. That taught me responsibility, scale, edge cases, and operational consequence, so I know what kind of reality AI has to enter.
Organization names are anonymized. Scope, responsibility, and outcomes are what matter here.Organization names are intentionally anonymized for a public-facing site. What remains is the part that matters more: scope, responsibility, and outcomes.
Built systems, growth loops, and cross-market execution inside a high-constraint travel business.Inbound Travel / Hotel Business Unit. Co-led group-level systems across 6+ key markets (JP/KR/MY/HK/SG/TH).
Planned national logistics infrastructure and governance to move new businesses from MVP into scale.Local Services Fulfillment Center. Planned national logistics infrastructure and agency governance, helping innovative businesses move from MVP into scaled operations.
Led a MaaS startup from product architecture to commercialization.MaaS innovation platform. Led the full loop from proof of concept to commercialization and secured seed-funding interest of over 10M RMB.
Optimized supplier performance and dispatch logic across a large on-demand delivery network.Instant delivery network. Built a multi-dimensional supplier performance matrix and optimized dispatch across 3,000+ delivery units.
Built the systems-thinking base through data mining and BI decision architecture.Recipient of the Merit Scholarship for Academic Excellence. Focused on data mining and BI decision architecture as the base layer for systems thinking.
If you want more than a chat box and need an AI interface that can enter the workflow, we should talk.If you want to turn AI from a demo into something that actually works — your own product, an internal tool, or a protocol layer — let's talk.
