Hobbyist developer working on computer networking, serverless edge deployments and self-hosted services.
About
I am a hobbyist developer who learns by building: DNS infrastructure, proxy and tunnelling stacks (Xray-core, sing-box), serverless deployments on Cloudflare's edge, and small automation around free-tier platforms. This site is my developer portfolio and lab notebook — the projects below are things I actually deploy, run and maintain. Everything here is personal, open and non-commercial.
Tools I work with: sing-box · Xray-core · Cloudflare Workers/Pages · GitHub Actions · DNS & TLS · Linux
Projects
Edge proxy nodes on Cloudflare Pages
I deploy and maintain VLESS/Trojan-over-WebSocket proxy nodes on Cloudflare Pages, based on the open-source Cloudflare-vless-trojan project — both TLS and non-TLS variants, with custom domains attached via CNAME + TXT verification, and a sing-box client configuration (rule sets, DNS splitting, urltest groups) on the receiving end. My running notes cover what breaks, what the platform's risk controls block, and what finally works.
instabox-keepalive — GitHub Actions automation
A scheduled GitHub Actions workflow that pings a free-tier VM every few minutes so the platform does not put it to sleep. The interesting parts are the scheduling trade-offs: polite intervals, idempotent pings, and accepting that free platforms have house rules. Code: github.com/chenshiyuani/instabox-keepalive
Free domain & DNS lab
I run a small portfolio of free domains and document how each registry behaves: EU.org (free forever, volunteer review, full nameserver delegation to Cloudflare), DigitalPlat domains such as dpdns.org and qzz.io (free renewal windows before expiry), KataBump subdomains, and developer subdomains like is-a.dev. The notes track which providers allow NS delegation, which only allow simple A/AAAA/CNAME/TXT/MX records, and how I keep renewals from lapsing.
After running sing-box, Xray and mihomo based clients side by side, I kept a single sing-box client in eBPF root mode. One well-understood tool beats three half-configured ones. Notes on rule sets, DNS splitting and urltest behaviour live here.
Reading a platform's failure modes
Free tiers fail in characteristic ways: an uptime test that reports every node dead while real traffic flows fine, a login blocked by datacenter-IP risk controls rather than a wrong password, a build pipeline disabled while the deployed site keeps serving. Half of self-hosting on free infrastructure is learning to tell "actually broken" apart from "the dashboard says broken".