Practical AI tools, workflow automation, self-hosted infrastructure, and other such nonsense.
I work in SEO and data-heavy problem solving. This is where I write about the small tools I build, the systems I run, and the technology claims that hold up once they reach a real workflow.
Start with the latest posts, or read more about PixelRaider.
Prompt engineering acquired the usual accessories of a new consulting category: courses, certifications, frameworks, and long lists of phrases that supposedly unlock better model behavior.
Most people need a better brief.
The model needs to know the job, the source material, the constraints, the shape of the answer, and how success will be judged. Contractors, employees, and software teams need the same information. AI makes a vague brief fail in seconds instead of weeks.
I have sat through hundreds of hours of analysis sessions. The number of those hours that changed what anyone did would round to zero.
The usual version starts with a reasonable question: “What percentage of our organic traffic was brand versus non-brand last month?”
GA4 says one thing. Search Console says another. The paid team has a third definition. Forty-five minutes disappear into source differences, query classification, and caveats. Eventually the group lands on 62% brand traffic.
A VPS backup earns its keep when the original server is unavailable. Provider snapshots help with routine recovery, but they remain inside the account and infrastructure you may be trying to escape.
This setup sends encrypted, deduplicated backups to Cloudflare R2. It covers web files, database dumps, selected system configuration, retention, integrity checks, and a staged restore process.
The goal is simple: a fresh Ubuntu server plus credentials stored somewhere outside the VPS should be enough to recover.
I wanted a tool that could ingest a sitemap and chart how many pages a site published each month. The output would show a prospect’s content velocity, or a competitor’s, without the usual spreadsheet cleanup.
In the past, I would have searched GitHub for something close, decided the setup looked annoying, and added the idea to a mental list. This time I described the workflow to Claude, worked through a few bad iterations, and had a usable tool about an hour later.