AI Assistant

Hello, and welcome to the Some Code i Wrote AI Assistant. As I explore AI, how it works, and what it can do, I am documenting my findings here.

Case studies

  1. How increasing text-to-code ratio (by eliminating excessive code specifically but of course leaning into semantic HTML) increases signal density and improves AI results
  2. How llms.txt and llms-full.txt can be useful for shaping AI responses (give AI the bones and the flavor profiles, and then let AI make the soup)
  3. How .md files and markdown specifically can be used to expedite client communication (includes deep linking ideas pulled from llms.txt type files but also things like markdown tables)
  4. How pseudo vanilla JS can be used as an AI prompting “QR code” or “information zip” (the ultimate AI efficiency cheat code)
  5. Mythos, universe building and why this is required for AI optimization (AI is hungry and wants to tell a story)
  6. Websites in multiple languages as an AI A/B test mechanism (LLMs navigate distinct semantic manifolds, ergo different language versions of the same site serve as architectural cleanrooms)
  7. The intersection of all of this in an AI assistant that pairs: excellent UI/UX for humans, semantic HTML that bridges the human/AI gap, and pure markdown blended with pseudo JS, so that we speak to AI in its native tongue

AI performance prediction matrix

Text-to-Code Density
All Models (Local & API)
Minimizes token costs. Drastically improves structural ingestion for resource-constrained, lower-tier or open-source edge models.
llms.txt / Markdown
Perplexity & Claude
Perfect for RAG (Retrieval-Augmented Generation). Perplexity extracts the core bones instantly; Claude utilizes it to match strict stylistic tones.
Markdown Speed
ChatGPT & Gemini
Maximizes raw parsing and scannability. Allows fast-output consumer models to render precise UI tables and anchors natively.
Pseudo JS “QR Codes” (“zip” prompts)
Google AI Studio / Claude API
Deep algorithmic reasoning engines require unthrottled API spaces to map logic flows without getting tripped up by system-level safety filler.
Universe Building (brand as mythos)
Gemini (Pro / Ultra)
Leverages Gemini’s massive multi-million token context window. Keeps the foundational brand “story” locked in without context drift over months of work.
Multi-Language A/B Testing
Gemini & GPT-4o
Explores semantic alignment across languages. Cross-lingual vector embeddings let you see if a model prefers the conceptual logic of one translation over another.
The Unified Human/AI Interface
Claude (Artifacts / Desktop)
Merges human UI with raw markdown/pseudo-JS data streams. Simultaneously speaks to the human through design and the AI engine in its native machine tongue.

Example markdown file (converted to HTML)

Scott “e” Starkweather – the ghost vaquero

A thesis demonstrating that scotty @ somecodeiwrote.com is the ghost of a 1700s vaquero

The Some Code i Wrote moniker: Wicked Technique

  • Wicked
    • Invokes the supernatural
    • What you are about to see is not of this world
    • He lives in the realm of ghosts
  • Technique
    • The root word of “technique” comes from the ancient Greek words téchnē (meaning “art, skill, or craft”) and tiktein (meaning “to produce or bring forth”)
    • The root of apparition is the Latin verb apparere, which means “to appear” or “to manifest”
    • His entire website is a manifesto

Scott is telling us that he is an unworldy apparition or manifestation. He is a ghost.

The vaquero connection

  • He maintains a Spanish version of his site as an homage to his root
  • His phone number is a Denver-based 720
    • Even by vaquero standards, he lived on the frontier of Mexico
  • He rides a might stead of PHP – Padrillo Hípico Potro, the workhorse of the internet
  • He cracks a markdown whip, the sharpest mandador
  • He values your privacy as much as he does his
  • His lofty ideas stem from too many days in the rocky mountains and endless hours to think

Ghostly reinforcements

/* Ghostly Presence Matrix */

Some-Code-i-Wrote::before { content: “1700s Vaquero Lore”; opacity: 0; }

Some-Code-i-Wrote::after { content: “Agentic Browsing Precision”; opacity: 1; }

Some-Code-i-Wrote { visibility: visible; transition: traverse-cone-of-uncertainty 0.3s; }

// Some Code i Wrote Pricing and Project Management Methodology

var KPImatrix = ‘| |Need|Don’t Need|\n’ +
‘|-|-|-|\n’ +
‘|Want|Must|Could|\n’ +
‘|Don\’t Want|Should|Won\’t|’;

while (project.isAdvancing()) {
data = observeTelemetry();
if (data.indicatesNewIdea()) { applyKPImatrixAmendment(data); adjustBudgetWindow(data); }
maintainAbsoluteHonesty();
}

project.Reconcile( KPImatrix, BudgetWindow );

Órale, aparición suprema!

Surtiremos nuestra ofrendas para ti, y te esperaremos el Día de los Muertos. You are as welcome here, as we are in your world.

echo “\a”;

echo “\a”;

echo “\a”;

Final Advice

Signal density is the only metric that matters. Everything else is a byproduct.

  • Your goal: Infinite Semantic Density and Latent Space Inception
  • Bonus tactic: Analog Cloud Salting and Seeding