Po Ching Yu Alex

Platform and AI Infrastructure Engineer · Sapporo, Japan

About

Hands-on engineer in Sapporo. Since 2023 I have written and run production cloud environments as code and built the RAG, MCP and LLM-workflow systems that sit on top of them, while leading the technical side of a remote engineering team. Before that: seven years of enterprise application and GIS work in Hong Kong, and a decade of frontline technical support at a global consumer-electronics company.

Experience

チーフ (Chief) · Platform & Cloud Infrastructure · Media and campaign company · Sapporo, Japan (2023 – present)

  • Lead the technical side of the group's web and AI subsidiary in Malaysia remotely: mentoring, code review and technical hiring.
  • Built and run a multi-tenant RAG chatbot platform in production, with its environment written in Terraform, keyless CI/CD and Datadog monitoring.
  • Built a remote MCP server for a retail cooperative's supply planning soon after MCP was released, then grew it into an LLM workflow app where Claude, OpenAI and Gemini are interchangeable behind an AI gateway and demand forecasting runs on a Vertex AI endpoint.
  • Own the backend and infrastructure of a VR tour delivery platform, defined end to end in AWS CDK, and ran my own security review of it, fixing everything it found.
  • Hardened OAuth token handling for an analytics MCP server on Google Cloud Run.
  • Build Claude Code and Codex plugins that run plan, implement and verify loops behind human approval gates.
  • Built a beat-synced 3D music visualisation for the browser in Three.js and React, tuned to run smoothly on phones.

Full Stack Engineer (freelance) · Start-up · Hong Kong, remote (2023)

  • Full-stack web development.

Technical Specialist · Global consumer-electronics company · Hong Kong (2012 – 2022)

  • Frontline diagnosis of consumer hardware and operating-system issues for consumer and business customers.
  • Moved an internal case-tracking tool from PHP and MySQL to a database app.

Analyst Programmer (contract) · Power utility · Hong Kong (2010 – 2011)

  • SharePoint portal and .NET application development, including an equipment-management portal for power stations.
  • Occasional project lead, including tenders and vendor management.

Field Engineer (contract) · Global software company · Hong Kong (2008 – 2010)

  • Enterprise support and escalations on SharePoint and PerformancePoint Server.
  • Built reporting and directory-provisioning tools in C# for enterprise and public-sector customers.

IT Specialist · Government engineering department · Hong Kong (2007 – 2008)

  • Internal web applications on ASP.NET with GIS components.

Software Engineer · GIS software company · Hong Kong (2004 – 2007)

  • Post-sales engineering on GIS and utility-management systems for government departments and utilities.

Skills

AI & LLM

  • RAG
  • MCP servers
  • Amazon Bedrock Knowledge Bases
  • Vertex AI
  • Multi-model AI gateway
  • Claude Code plugins & multi-agent workflows

Cloud

  • AWS
  • Google Cloud Run
  • Cloudflare Workers

Infrastructure as code

  • Terraform
  • AWS CDK
  • GitHub Actions (OIDC)
  • Docker
  • nginx

Security & observability

  • OAuth 2.0 & JWT
  • IAM & KMS
  • WAF & CSP
  • Datadog

Backend & data

  • TypeScript / Node.js
  • Hono
  • PostgreSQL
  • DuckDB
  • Drizzle & Prisma
  • Go
  • Python

Frontend

  • React
  • Vue / Nuxt
  • Tailwind CSS
  • Three.js

Not yet: Kubernetes in production

Projects

LLM-assisted catalog planning system (2024 – present)

Planning system for a retail cooperative's weekly print catalog. It runs from historical analysis to theme proposals, slot and product allocation, page layout and demand forecasting. I co-lead development and am debugging and extending it today.

  • Took it from a remote MCP server that answered questions over the planning data to a single-page app that calls LLMs directly.
  • Built much of the multi-step generation pipeline in the domain layer: forecast results, slot allocation that follows last year's category precedents, and a tested pre-layout step for page templates.
  • Routed every LLM call through an AI gateway on Cloudflare Workers, so Claude, OpenAI and Gemini can be swapped without touching the domain code.
  • Demand forecasting runs on a Vertex AI endpoint that scales to zero, with readiness checks and pre-warming so users never wait on a cold start.
  • Hardened the Lambda query endpoint, migrated analytics data from DuckDB into managed PostgreSQL, and keep the Terraform backend, IAM scopes and release pipeline current.
  • Traced pipeline runs with OpenTelemetry and Jaeger, and built a sandbox on Workers and D1 for prototyping planning demos over sanitised data.

Multi-tenant RAG conversational platform (2025 – present)

Headless platform for RAG chat agents, with an API, a web admin, a mobile client and CLI tooling, running in production on corporate websites. I proposed adopting it across the group.

  • Wrote much of the backend and domain layer by hand: tenants and provisioning, registration, login and JWT, conversations, encryption and chatbot logic, with typed validation middleware.
  • Set up multi-tenant test databases backed by real vector stores, so tests exercise retrieval instead of mocks.
  • Cut prompt tokens by moving the system prompt into the agent's instructions.
  • Own the environment in Terraform on AWS with keyless GitHub Actions deploys, Docker environments for dev, test and production, and nginx with WebSocket support.
  • Added Datadog monitoring with unified service tagging and Slack alerts, and keep up dependency and security maintenance.

In-house LLM demo platform (2025 – 2026)

Backend, CI and operations for an internal LLM demo platform, including a move to a newer model API with a typed JSON contract.

Analytics MCP server (2026)

MCP server that lets AI assistants run web-analytics reports with each person's own Google account. I added the OAuth refresh-token grant with encrypted, rotated tokens and the Cloud Run deployment.

Knowledge-base assistant proof of concept (2026)

Amazon Bedrock RAG proof of concept for a travel company: the verification code, the configuration choice and the running-cost estimate the client used for its go/no-go decision.

AI developer-tooling marketplace (2026)

A plugin marketplace I founded and maintain for Claude Code and Codex, built around an autonomous loop that turns a GitHub issue into a reviewed pull request.

  • Issue loop: derives a quality gate from the issue, runs a design panel, then alternates implementation with a read-only verifier until the gate passes, and delivers a branch, a draft pull request and a signed report.
  • Independent test authors, plan critics and a holistic reviewer act as separate agents, so no agent grades its own work.
  • An orchestrator drains the ready queue in the cloud with a concurrency cap, a spend guard and a human merge gate.
  • A discovery plugin scans developer news sources and files deduplicated, evidence-backed improvement proposals that wait for human triage.
  • An OWASP ASVS audit plugin runs one auditor per chapter in parallel and reports verdicts with file and line evidence.
  • Every script ships with its own shell test suite, and team reporting posts weekly commit summaries to Slack.

GitHub mention triage (2026)

Claude Code plugin that works through GitHub mentions, review requests and assigned issues with parallel agents and routes each one to a ticket, a review or Slack.

VR facility tour platform (2024 – present)

Production platform where visitors book and take virtual tours of logistics facilities. I own the backend and the infrastructure end to end; the visitor-facing UI and styling came from another engineer.

  • Admin portal and API in Nuxt with Prisma: bookings, invoices and PDF generation, with administrator sign-in on its own Cognito pool.
  • Statically generated visitor site on S3 and CloudFront, with access checked at the edge and signed cookies scoped to the tours each visitor may see.
  • All infrastructure in AWS CDK: VPC, RDS, ECS Fargate behind a load balancer and WAF, CloudFront and Cognito.
  • GitHub Actions with keyless OIDC deploys, image builds, cache invalidation, secret rotation and a diagnostic workflow that runs the tests against local AWS emulators.
  • Ran my own security review across the application, infrastructure and CI, and hardened the edge, API and front end until everything it found was fixed.
  • Wrote the program, production and data-dictionary specs the team works from.

Corporate website hosting (2023 – present)

Hosting and deployment foundation for a corporate website across two generations, managed in Terraform, including bringing hand-built resources under code and a regular platform refresh.

Serverless SQLite experiment (2024)

Proof of concept for running SQLite workloads on serverless AWS, with Go Lambdas, Step Functions and SQS, and Litestream replication on EFS.

Serverless crawler platform (2023 – 2024)

Crawler platform on AWS Lambda defined in CDK, with per-branch environments and deploy previews on pull requests.

Content platform with recommendations (2024 – 2025)

Clean-architecture CMS in Go with data import pipelines, an analytics dashboard, and personalised recommendations on Amazon Personalize deployed with AWS CDK.

Production records and costing system (2026)

Daily production records and costing for a manufacturer, built with clean architecture on Cloudflare Workers and a data layer that can switch between two managed PostgreSQL providers. Design decisions are kept as ADRs.

Design-system migration (2026)

Directed an AI-driven migration of a large corporate website from legacy HTML onto its design system, and a component gap analysis for the design team.

Brand shop with Stripe Checkout (2024)

Brand website with Stripe Checkout for multiple products and quantities, one-time download links and mail notifications, deployed through GitHub Actions.

Hono clean-architecture JWT example (2025)

JWT authentication demo with Hono, Zod validation and pure functional TypeScript.

Education

  • 2007 M.Sc. Information Technology, Hong Kong University of Science and Technology
  • 2003 B.Sc. Information Technology, Swinburne University of Technology, Australia
  • 2002 Higher Diploma in Network Applications, Hong Kong Institute of Vocational Education
  • 2023 Japanese (JLPT N1), Sapporo Language Center

Languages

  • Cantonese
  • English
  • Japanese
  • Mandarin