What's Under the Hood of This Site

Found by AI search

Built to be read by AI search, not just Google

People now ask ChatGPT, Claude, and Perplexity to go look things up for them.

So I set the site up so those tools can read it and describe me accurately, instead of guessing.

A plain file called llms.txt, which is a summary of the whole site written specifically for an AI to read.
Settings that welcome the AI readers instead of blocking them.
Structured data behind every page, a hidden description of who I am, what each page is, and how my profiles connect, in the format machines understand.
A finished copy of every post pre-built for those readers, since they do not run the code that builds a page in your browser, so they see the real writing and not an empty shell.
I also run a suite of GEO tools that read the site the way an AI does, scoring its structure, its schema, and how easy it is to quote, so I can fix what they flag instead of guessing.

Consistent imagery

The same person in every blog photo

This one started as a real problem.

When I wrote the post about helping my mother, I let the tool generate the images for it.

There was no character consistency at all, so nobody stayed the same from one photo to the next.

My mother was a completely different woman in every single image, and so was the daughter who was supposed to be me.

It made the whole post look thrown together.

So I took a course on AI video generation, and one of the things it taught was character consistency.

Then I trained Claude on what I learned, and built the system I use now.

I built a 62 image character kit, a locked set of reference photos of the same person from every angle, in different light, with different expressions.
An image pipeline uses that kit so she stays consistent across all of my posts, not just within one.
Each post still gets its own feel, so I change her hair, her clothing, and her setting to match the story.
Every image carries the same photographic recipe, so they look taken on one camera in natural light, instead of looking like random AI pictures.

My own voice

Every post is read aloud in my real voice

I did not want my posts read to you in a generic AI voice.

So I cloned my own voice and trained it to read them.

This was by far the hardest thing I built on this whole site.

Honestly, I thought this one was going to be a piece of cake.

Almost everything else here came to me pretty easily, even the technical parts.

This one did not, because I was not expecting it to be hard.

It took more tweaking than anything else I built, and it is still the thing I am most particular about, because it is my voice.

When someone hears it, they form an immediate impression of who I am.

If I am not reading well, or if it sounds flat and emotionless, that becomes their picture of me.

I have a great sense of humor. I love having fun.

I have not been able to get that across yet - I'm hoping to one day.

I really want people to get a sense of who I am, so I take it personally when the cloned voice is shrieking the words out 😂.

The training recording had a faint hum from an air purifier (whoops....lol), so I built a filter that strips it out of every narration automatically.
The voice would stumble on odd words and numbers, so I built a way to tell it exactly how to say something without changing what you read on the page.
A scanner reads each post before it is ever recorded and flags anything that would trip the voice up.
When a clip sounds off, a tool measures it and tells me what is actually wrong, instead of me guessing by ear.

I wrote about the whole saga here

Read-along

Follow along as it reads to you

I did not want the voice to just read at you while you stared at a wall of text.

I wanted you to always know exactly where you are on the page, so I built a read-along that highlights each line as it is spoken and gently scrolls the page to keep up.

The player controls also follow you down the page with play, pause, rewind, speed control, and a volume slider, so you always have easy access to those.

It is a separate piece of work from the voice itself, and keeping the highlight perfectly in time with the audio was its own project.

In-page blog editor

Editing without burning through tokens

Every change to my blogs used to mean asking Claude to make the edit for me.

For small fixes, a comma here, a reworded sentence there, that added up and burned a lot of tokens fast.

And asking Claude for a small edit is one of the most expensive things I can do in Claude Code. Every change is a full round trip: Claude has to read the whole post, work through it, and write it all back, even when all I changed was a single word.

So I built an editor right into the page.

Now I can open any draft, click into it, make small changes myself, and publish it when I feel it is ready.

I can fix a word, change a title, swap out the images, or hyperlink anything, without paying for a full back and forth every time.

The writing pipeline

How the posts actually get written now

Claude was writing my posts in my voice, and doing it well.

But as I mentioned, having Claude do that work is one of the most expensive things I can have it do for me.

And it costs only pennies to have Gemini do the same thing instead.

So once Claude had my written voice down so well, it wrote the instructions for Gemini on how to do it.

Now Gemini writes the posts instead.

I give Claude the outline, and it hands the writing off to Gemini.

This has only been happening for a post or two, so I do not have a solid read on it yet, but it seems to be working out great.

Gemini writes the post in my voice, formats the whole page, adds the images, and breaks everything into one sentence per line the way I like it, so it is easy to read.

And now, with the built-in editor, I can make any small tweaks on my own.

Then I publish it, and Claude adds the narration in my cloned voice.

Underneath all of it, there is a written set of rules for exactly how my written voice sounds, plus automatic checks that catch a draft slipping out of my voice.

The projects

And the projects it shows

None of these started as portfolio projects. Each one came from a real gap: something I needed, something a friend needed, or something a team needed and couldn't get from off-the-shelf tools. Every one is a real working tool, and most ship with the same layers around them: a showcase page, a live demo you can click into, a guided tour, and a walkthrough narrated in my voice.

Recipe App

My recipes were everywhere: handwritten cards, folders, Notion, and I had no good way to decide what to make from what I had on hand. So I built one. It reads a handwritten card from a photo, imports from a link, takes a voice memo, and lets you talk to an AI to figure out dinner from whatever is in the house. 215 real recipes in production.

See it ›

Friends Job Dashboard

Three friends from my last company were laid off and losing hours a day on job boards, getting nowhere. So I built them a private dashboard. It scrapes matching jobs, scores each one for fit, learns what each person likes, and tailors a resume and cover letter to a specific job with one click.

See it ›

OnTrak Call Guide

OnTrak had over a hundred specialists making member activation calls with no shared call structure, so the ones who struggled had nothing to follow. So I built a call guide that gave every agent the same foundation from day one: six stages, every call, with scripts for different health conditions and a narrated walkthrough.

See it ›

Life at a Glance

Running more than twenty active projects across different AI tools, I had no way to see the state of everything without opening each one. So I built a dashboard that surfaces what needs attention across all of it, grouped by area of life, and flags what would otherwise slip through.

See it ›

Visual Prompt Builder

I was studying AI image and video generation, and it felt like learning another language, a vocabulary most people are never taught. So I built a tool that handles the translation: plain English in, a professional prompt out, and it can work backwards from images I like.

See it ›

Entertainment Hub

On a Friday night I could never remember which series I was in the middle of or what platform it was on, and recommendations from my sisters had nowhere to go. So I built one app that tracks everything across every service, with an assistant that knows the whole library.

See it ›

And every one of those projects opens with a short motion-graphic intro video, each narrated in my own cloned voice.