I ship the product, run the servers, and grow the audience.

Designed and built the Komunitech course platform end-to-end

The front-end design and the source code: a public landing page, a member dashboard with a catalogue of AI models and learning paths, a course player for markdown, video, PDF and quiz lessons, and an admin panel with a curriculum builder for sections, lessons and quizzes, plus pricing, voucher, roles and payment modules.

How it runs. Vue 3 and TypeScript on Vite and Bun, with type-checked builds; authentication through Clerk; key routes prerendered for search engines; GitHub Actions deploys over SSH to the VPS and syncs the build into place. Courses can be assembled from a bundle format (classes, sections, lessons, quizzes) so a new class goes live without touching code.

Rebuilt the blog as a standalone WordPress theme

A theme written from scratch to replace the page-builder layout: a seven-section homepage, a three-column article grid, article pages with an automatically generated table of contents, category pages and a four-column footer. Shipped as v1.0.4 and serving every post on the blog.

What it fixed. The table of contents used to overlap content on phones and lose its sticky position; both are solved. Minified-CSS caching that kept visitors on stale styles after a release is handled with versioned assets.

An automated pipeline that turns 24 hours of AI news into ranked articles

Twice a day the pipeline collects the last day of AI news, drafts an article to the Komunitech editorial SOP (answer-first opening, key takeaways, at least six sections, FAQ, references, internal links), places the focus keyword where Rank Math expects it, and publishes with metadata and a banner.

Quality gate. A Python audit script checks each draft for structure, forbidden phrasing, link rules and keyword placement before it can go out. When early articles scored in the low 20s on Rank Math, tightening keyword placement moved the same articles to 87 out of 100; the target is 80 or higher for every post.

Daily newsletter delivery, from template to send-ready HTML

HTML email templates filled from the day's content, with every image uploaded to a CDN (Bunny, through its API) so editions render fast in any mail client. The output is final, send-ready HTML for each daily edition, with nothing left to assemble by hand.

The VPS everything above runs on

A Contabo VPS (4 vCPU, 7.8 GB RAM) hosting more than ten domains and services behind nginx, with Let's Encrypt certificates, DNS on Namecheap and Tailscale for private networking between machines.

Deploy discipline. Every service has a written runbook precise enough for an AI agent to follow, releases go through CI and rsync into versioned directories, and a change is only considered done once the live page has been checked.

Automation that publishes, assembles courses and uploads to the CDN on its own

AI agents given the SOPs, scripts and credentials they need to do recurring work without supervision: publishing articles, assembling course material into the platform's bundle format, and pushing imagery to the CDN. Each job is scheduled and ends by verifying its result on the live surface, so a silent failure cannot go unnoticed.

Reusable pieces. The editorial rules live in a skill package (instructions plus an audit script) that any agent can load, so writing quality does not depend on who, or what, is drafting.