AI-Powered Tech, Growth & Business Operator
Most people do one side or the other. I do both, which is the only reason one person covers the range.
I'm an operator, not an ML engineer. I deploy and orchestrate AI, I don't train models. I run AI systems the way a director runs a set: the value isn't typing every line, it's knowing what to build, directing the tool to build it right, and catching what's wrong. See the method →
Every build here is real, with an honest role and a live link where the work is public. Open any one for the full case: problem, what I did, the result, and what I caught the AI getting wrong.

Full brand change and a deeper marketing site for Paralegent.ai, shipped continuously on live main without ever taking it down.

cognilium.ai, live: 76 routes and a 22-post clustered blog on a CMS where you publish once and the page is structured, optimized, and indexed automatically.
Found 19 of 46 pages silently un-indexed, recovered them, and left behind a self-maintaining SEO/GEO system plus a citation moat for AI answer engines.
Stood up cognilium.ai/tech-news as a real newsroom that qualifies for Google News and Discover: NewsArticle schema, news-spec sitemaps, RSS/Atom/JSON feeds, and Publisher Center registration.
One publish in the CMS now auto-syndicates a post across six channels off a single webhook, idempotent per slug so the same article never double-posts.
Six agent-skills auto-enforce the locked palette, voice, page rhythm, FAQ rules, and conversion psychology on every change, across Cursor, Codex, Gemini, and other AI coding agents.
Reverse-engineered a competitor's API to lift its entire 651-product catalog into a clean, audited dataset, every record carrying a SHA256 receipt of where it came from. The full dataset is downloadable below.
Designed the 16-table database and built the directory that serves it: 575 products scored by a transparent model where sponsorship never touches ranking, on a ~2,000-page Next.js site.
Cinematic AI marketing video with voice-cloned narration and AI visuals, scripted and edited end to end for a paying UK client.
This is the part most people hide. I lead with it, because it's the thing that's hard to copy.
I decide what to build and what 'good' looks like, the call AI can't make. Brand direction, which approach wins, what to cut. The taste is mine.
I drive AI across a real server and codebase: context management, breaking work into what the tool can actually execute. Most people get toys out of these tools. I get production.
I author agent skills and rule files that enforce brand, vocabulary, page rhythm, and quality on every change, so the output stays correct at speed and across tools.
I review, test, catch the mistakes, and ship on live main. The output is only as good as the operator's ability to judge it. That judgment is what I bring.
The honest way to judge me is the work itself. Every case study shows the call I made and what I caught the AI getting wrong.