Three months ago I had a portfolio site, a handful of school projects, and an idea for making my family’s Secret Santa less chaotic.

Today KovertKlaus is a live, database-driven web application. It runs on a domain I own, people outside my house use it, and when one of them reported a bug this afternoon, the fix reached production before dinner.

Not many people know it exists yet. It is still public, it still works, and I built it. This post looks back at how that happened.


The Timeline

DateMilestone
Jul 26Published this site, an Astro portfolio on Cloudflare
Jul 29npx create-next-app, the first KovertKlaus commit
Aug 2Derangement-based draw engine, dashboard, Command Center, and a first round of OWASP hardening
Aug 13Opened the source repository to the public
Aug 18kovertklaus.com went live on Cloudflare
Aug 20Moved to a full-stack Cloudflare Worker backed by serverless PostgreSQL, plus an admin console and an internal QA lab
Aug 20 – Oct 6Seven weekly sprints of six deliverables each, with a retrospective every Monday
Oct 1Tagged v0.2.0-alpha, on the date I committed to in August
Oct 12Automated the dev-to-production sync pipeline
Oct 13First unprompted praise from a user, and the first user-reported bug fixed in production the same day

That is 76 days from an empty scaffold to a product with users.


By the Numbers

  • ~40,000 lines of TypeScript across 164 source files
  • 18 PostgreSQL data models managed with Prisma
  • 240+ commits in the development repository
  • 201 / 201 automated tests passing
  • 0 known npm vulnerabilities at launch, across 525 packages
  • 1 release on time: v0.2.0-alpha on October 1

Line counts are not a measure of quality. I list them to show the scale of what one person on nights and weekends can carry with good tooling and discipline.


What Actually Mattered

1. Shipping early to a real domain

KovertKlaus went public on August 18, three weeks after the first commit and long before it was “done.” It served a pre-launch page at first, but production was real: TLS, security headers, DNS, a CI/CD pipeline, and a database. Every later feature landed in a working deployment, not in a pile of work waiting for a launch.

2. A cadence I could keep

Each week got exactly six deliverables, and each Monday started with a short retrospective. Six is small enough to finish next to a full-time job, school, and family, and large enough to move every week. I hit the alpha date I set in August because of that rhythm, not because of any heroics.

3. Security as a baseline, not a phase

Session authorization checks, input sanitization, SSRF guards on the link scraper, strict security headers, separate least-privilege database roles for the public app and the admin console, and dependency overrides to clear upstream CVEs all went in early. Twenty years of keeping systems up taught me that hardening added later rarely gets done.

4. Building the delivery loop, not just the features

The milestone I am proudest of is not a feature. It is the loop:

user report → diagnosis → fix branch with tests → pull request → automated production sync

That loop ran for real on October 13. I wrote about it in KovertKlaus Alpha: Surviving the First Bug Report.


Everything Else Along the Way

KovertKlaus got most of my attention, but the quarter also produced:

  • A cross-platform workstation profile. One set of shell, terminal, and git configuration for my Linux Mint and Windows 11 machines.
  • A terminal XKCD viewer. High-resolution comics rendered inside the terminal, aligned to the character grid, and shown in my login banner.
  • A subnetting teaching method. Visualizing Subnetting: The 4B Method, written for students who struggle with VLSM.
  • An AI pair-programming workflow. A versioned, encrypted knowledge base and a set of rules that let an AI assistant pick up work across four machines without losing context. It is a large part of why the pace above was possible.

What I Learned

  • Contracts drift silently. The first production bug was a mismatch between what the browser sent and what the server expected. Shared, tested definitions are cheaper than debugging (details in the follow-up post).
  • Edge runtimes are not Node. Moving to Cloudflare Workers meant rethinking database connections, file access, and logging. Serverless PostgreSQL with a pooled connection string and a direct one for migrations solved most of it.
  • Dates create focus. Setting the alpha date publicly in August shaped every sprint after it.
  • Real users beat any test suite. Within hours, alpha users found a bug that 195 passing tests had missed.

What’s Next

  • v1.0.0-beta on November 1: the Season 1 public winter launch.
  • Capacity checks before beta: email sending limits, database headroom, and load behavior.
  • Listening to early users: the people who are excited now are the best guide to what to build next.

If you organize a Secret Santa or White Elephant exchange this year, I would love for you to try it at kovertklaus.com. If you would rather run it yourself, the source is on GitHub and free to self-host for personal and non-commercial use.