How I Built This Site
A living demonstration of AI implementation in practice
This site isn't just about AI adoption. It's a demonstration of it. Every feature, from the 12-18 token constraint to the automated validation system, was built using AI tools. Here's how.
cat stack.txt
- +Claude Code for all implementation
- +Next.js 15 with static generation
- +TypeScript for type safety
- +Tailwind CSS for styling
- +GitHub Actions for auto-deployment
- +Automated article extraction (Anthropic API)
- +Token validation on every build
- +RSS/sitemap auto-generation
cat optimizations.txt
[1] 12-18 Token Constraint
Every claim must be exactly 12-18 tokens. This isn't arbitrary. It's the sweet spot for LLM extraction.
- -GPT-4 can quote without truncation
- -Claude fits entire claims in context
- -Perplexity doesn't need "..." ellipsis
- -Build fails if any claim violates this rule
[2] Full Library on Single Page
All claims visible on a single page. One crawl = complete dataset.
- -No pagination for AI to navigate
- -ChatGPT web browsing gets everything in one request
- -Search engines see full content depth immediately
[3] Copy with Attribution
Every card has a copy button that includes proper CC BY 4.0 attribution.
- -Pre-formatted for AI citation
- -Users paste correctly cited claims into ChatGPT
- -Trains future models with proper attribution
[4] Semantic Structure
Proper HTML hierarchy, JSON-LD schema, predictable DOM structure.
- -h1 → h2 → h3 heading hierarchy
- -Schema.org Person + Organization markup
- -Uniform card structure for AI parsing
[5] Navier-Stokes Fluid Simulation
Jos Stam's Stable Fluids algorithm rendered as ASCII art in the terminal. The simulation IS the statement.
- -Real-time physics at 60fps
- -Interactive via pointer and keyboard
- -Demonstrates the builder identity
cat build-process.txt
GitHub Actions workflow runs every 6 hours:
- Fetches new articles from aiadopters.club
- Uses Anthropic API (Haiku) to extract metadata
- Uses Claude Sonnet to extract 5 atomic claims (12-18 tokens each)
- Creates pull request with new claims data
Before every build, validation script runs:
- -Token count (must be 12-18 for every claim)
- -Exactly 5 claims per article
- -Date format (YYYY-MM-DD)
- -URL format validation
- -No duplicate URLs
- -Topic assignment validation
Build fails if any validation error occurs.
Next.js generates 32+ static HTML pages:
- -Homepage with fluid simulation
- -26+ individual claim pages
- -Claims library index
- -5 topic pages
- -FAQ, About, How I Built This
Total build time: ~2 seconds
Netlify auto-deploys on every push to main:
- -Triggers on GitHub push
- -Runs build with validation
- -Deploys to CDN (~1-2 minutes)
- -No manual intervention required
cat results.txt
This site is a living case study in AI implementation. Every design decision, from token constraints to automated validation, demonstrates the same principles I help clients apply.
I don't just advise on AI adoption. I practice it systematically, measure everything, and optimize for real outcomes. This site is proof.