How This Site Works

01 · AI Assistant

The terminal on the homepage

The terminal is a real AI assistant powered by Claude, running behind a server endpoint on this site. It answers only from a hand-written knowledge base covering every page: experience, results, playbooks, blueprints, and tools. That grounding keeps it factual instead of letting the model improvise.

It also adapts quietly. The visitor context the site already has (your rough location, or a company name when one is known) is passed to the model with each message, with instructions to use it subtly and never announce it. A visitor from a fintech gets fintech-flavored examples without being told "I see you work at a fintech."

02 · Personalization

The greeting that knows where you are

The hero greeting composes itself from three signals, in order of confidence:

• URL parameters: outreach links can carry ?name= and ?company=, so a message I send you opens with your name already in the greeting.
• IP-to-company lookup: a server endpoint resolves office IPs to a company name, gated on a confidence score with an ISP deny-list, and cached so repeat visits cost nothing. Home and mobile connections stay anonymous.
• Geolocation: a local-language hello for your city or country. City names only appear for recognizable metros; a suburb broadens to the country so it never reads wrong.

03 · AI Search

Built to be read by machines

AI answer engines are becoming a first-touch channel, so the site treats them as an audience:

• llms.txt and llms-full.txt: plain-text maps of the whole site following the llmstxt.org spec, generated from the same data files as the rendered pages so they never go stale.
• Ask-AI footer links: one click opens ChatGPT, Claude, Perplexity, Grok, or Google AI Mode preloaded with a prompt pointing the model at these pages, so its summary is grounded on real content.
• Structured data: every page ships JSON-LD (person, breadcrumbs, articles, collections) so both search engines and LLMs can parse who did what.

04 · Generated Assets

Pages that build their own artifacts

Anything that could go stale is generated from the same data at build time:

• Social preview images: every page gets its own OG image rendered from its title, with the text placed inside the square crop zone so link previews never cut it off.
• The CV: the printable CV at /cv renders from the same data file as the experience page. Update the data once, both update.
• Sitemap and feeds: sitemap, RSS, and the llms.txt files all derive from the content data, not from manual lists.

05 · The Stack

What it runs on

Astro with React islands, Tailwind, and static output on Vercel. The AI assistant and the company-reveal lookup run as server endpoints; everything else is prebuilt HTML. Cloudflare handles DNS, a strict Content-Security-Policy allowlists every external script, and analytics run through Google Tag Manager.

The site is designed, written, and maintained with AI coding agents doing the heavy lifting: the same way I build growth systems. This page exists because the site itself is the best demo of that work.