Personal CMS

The problem

I write one or two articles a week, and a newsletter, Both Hands on the Wheel, about every two weeks that wraps them up. The articles cover practical AI at home, enterprise AI from the inside, and the craft of writing with AI. Each piece goes to LinkedIn, and each one needs a cover image, a promo post and a check against my style rules. After that, I want to see how each piece performs.

Done by hand, that's a lot of separate steps on separate sites every week, and any one of them is easy to drop. A correction I made to one draft could also slip back into the next.

What I built

A web app that runs on my own computer. Every piece gets drafted, checked, previewed, published and measured in one place, with my pipeline and style guide right there too.

• Articles: every draft and published piece with its cover. I preview each one exactly as it will look live, review it, then publish to LinkedIn, or push an update after an edit.
• Performance: impressions, reactions, comments, profile clicks and followers over time, by week, month or quarter, for each article and for the account overall.
• Engage Topics: finds like-minded people on LinkedIn and good conversations to join, on the topics I care about.
• Research List: things I want to learn, try or explore. My other projects can add to it too.
• Roadmap: my running list of article ideas and other work I'm planning.

The automations behind it

• Drafting: every article starts with a long voice recording of me talking through my ideas and points. AI makes a rough first draft from that transcript, using my curated style guide. Before I see it, a script checks it against my voice rules and my list of banned AI words, showing its evidence line by line. Then I keep revising until it's 100% my voice, my words and my ideas. Every correction I make becomes a hard rule, so a fix can't slip back in.
• Cover images: built-in Gemini image generation, with a prompt for each article. It can edit an existing cover instead of starting over, and it crops each image to LinkedIn's size.
• Publishing: it works through every step in LinkedIn's editor for me: pasting the formatted text, attaching the cover image, setting the newsletter and clicking through to publish, so headings, links and images all come through intact.

What had to be handled carefully

• Flawless publishing. What goes live has to be exactly my words and formatting, with every image and link intact. A piece only counts as published once the app confirms it landed on LinkedIn. If any step fails, it stops and leaves my draft untouched.
• Full transparency. I can see every step, what triggered it and what it changed, so nothing happens that I don't understand. When I edit a published piece, it flags that LinkedIn needs an update. Since the words have to be mine, nothing can be hidden or over-automated.
• Accurate numbers. Plain code collects and checks my stats, not AI, so what I see matches LinkedIn.
• The right conversations. Engage Topics keeps up with what I'm focused on now. It screens out AI slop and posts that don't add anything, and surfaces like-minded people I'd want to talk with. It also only shows posts recent enough that a comment still gets seen.

The value

Writing with AI takes care. It's an iterative process, and every step needs to be visible and easy to work with. That's where the risks of writing with AI get caught, before they reach anyone. The publishing and stats tracking are tedious, and doing them by hand would eat hours every week. Compared with a manual process or an off-the-shelf tool, this system is built around how I actually work with AI. The result is higher quality, authenticity and a big gain in productivity.