Case study
Full-stack store built from scratch and running live: checkout with stock reservation, card (3-D Secure) and PromptPay payments, shipping and tracking, loyalty points, real-time notifications, and an admin CMS — deployed by CI/CD to my HomeLab.
A full-stack e-commerce platform I built alone to learn how a real store fits together — storefront, API, data, payments, files, background jobs, and deployment — and then took it live. It runs at shop.flukelnwza.online on my own HomeLab behind Cloudflare Tunnel, and every push to main is tested in GitHub Actions and deployed automatically.
Commerce looks like CRUD but fails at the seams: two customers buying the last item, a payment that succeeds after the order expired, a webhook that arrives twice, a notification sent for a change that rolled back. The project exists to handle those cases deliberately instead of hoping they never happen.
Shoppers and shop staff (admins, moderators). It runs on a single home server with a demo catalog, so it is a working deployment rather than a business with customer volume.
Individual builder — designed, implemented, tested, and deployed the frontend, API, data model, and infrastructure.
Browser / Next.js Frontend
Storefront & admin
Express API
REST, auth, payment worker
Omise Gateway
Cards (3-D Secure) & PromptPay
PostgreSQL
Orders, stock, content
Redis
Sessions & rate limits
S3 Storage (SeaweedFS)
Object storage
The order and its stock reservation are committed first, then the card is charged. A lost Omise response is treated as unknown rather than failed, and a worker reconciles it, so money is never taken for an order that rolled back and a customer is never charged twice by a blind retry.
Notifications are inserted in the same transaction as the status change and announced with pg_notify, which PostgreSQL only sends on commit. A listener pushes them to sockets and email, so nothing is delivered for a change that was rolled back.
The server sits behind Cloudflare Tunnel with no inbound ports. GitHub only sends an HMAC-signed request; the server pulls the commit itself, checks it is on main, deploys, and rolls back if the health check fails.
Infrastructure
An old PC rebuilt as a Proxmox server that runs my live projects and a local LLM on its own GPU, published through Cloudflare Tunnel with Grafana and Dozzle for metrics and logs.
AI / ML
LiveSpeak to your phone and your Windows PC does it — with no cloud AI. Speech-to-text and an LLM run on my HomeLab GPU; a policy layer and an HMAC-signed agent decide what is allowed to run.
