Matthew Scott - Web Developer & AI Consultant, Louisville, Kentucky | Project Lavos

About Matthew Scott

Matthew Scott is a full-stack web developer and AI consultant based in Louisville, Kentucky (40216). He is the sole founder and operator of Project Lavos LLC, a Kentucky limited liability company (KY SOS filing #1545983.06, filed February 20, 2026). He specializes in React, JavaScript, Python, and AI integrations, building production-grade websites for small and medium-sized businesses. Known online as guitargnarr. His tagline: "Complexity, untangled. Turning ambiguity into systems that actually ship."

Contact: matthewdscott7@gmail.com | 502-345-0525 | LinkedIn | GitHub | Instagram

Louisville Client Portfolio (69 Deployed Demo Sites)

All sites are spec work — built by Matthew Scott from publicly available business information as proof of capability, not pitches. Each is live, deployed, and working on Vercel. Every site footer links back to projectlavos.com with "Built by Matthew Scott at Project Lavos." The full portfolio is browsable at projectlavos.com/#louisville with category filtering. Machine-readable portfolio data is available at portfolio-urls.json.

Portfolio categories with representative sites:

Louisville neighborhoods represented: St. Matthews, Germantown, Middletown, Jeffersontown, the Highlands, Dutchmans Lane, Downtown, Crestwood, and Westport Road.

Practical Applications (4 Deployed & Working)

  • FretVision — Interactive guitar learning platform combining fretboard visualization, a scale reference library of 24 PDF books (11,520 exercises), tab playback, and riff generation. All tablature is mathematically verified using Python. Stack: React, Python, MIDI, Canvas API, Supabase.
  • Meridian — US economic data explorer with interactive state-level maps. Visualizes unemployment rates, median household income, population, and poverty metrics for market intelligence. Stack: Next.js, TypeScript, D3.
  • Jobway — Full-stack job search automation platform. Gmail OAuth integration, NLP-powered ATS optimizer, smart follow-up scheduling, and analytics dashboard. Not a template — actual working automation with real API integrations. Stack: React, FastAPI, spaCy, Gmail API.
  • OurJourney — A private space for two people to document what matters, without an audience. Shared memories, private journal entries, and experience tracking. No followers, no likes, no public feed. Stack: React, Supabase, Vercel.

Conceptual Explorations (10 Interactive Experiences)

Interactive experiences exploring philosophy, psychology, and narrative through code. Built with React, Three.js, WebGL, Web Audio API, and p5.js.

Jungian Depth Psychology Series (6 pieces)

  • Psyche — Four portals into Jungian depth psychology: Shadow, Anima/Animus, Self, and Collective Unconscious.
  • Threshold — An exploration of liminal spaces and psychological transformation.
  • Umbra — Shadow work and integration.
  • Forma — Universal archetypal figures across cultures and mythology.
  • Auriga — Human-AI collaboration through Plato's chariot allegory.
  • Medittoica — Stoic philosophy as a 4,200-particle WebGL constellation.

Scroll-Driven 3D Experiences (4 pieces)

  • The Quiet Trade — Five philosophical scenes examining AI adoption in professional work. Procedural ambient audio responds to scroll position.
  • The Pattern Break — A personal narrative: 119 days from broken binary to business entity. Procedural wind and wind chime audio.
  • The Skeleton Crew — Hybrid WebGL and p5.js rendering with granular synthesis audio.
  • Regenera — Five acts of transformation with photographic textures and cinematic dissolve transitions.

Development Methodology

Four core principles guide all work at Project Lavos. These are detailed in the manifesto and explored further in The Verification Gap article.

  1. Substance first — Working software over impressive pitches. All 69 portfolio sites are deployed and working, not mockups or screenshots.
  2. Built to last — Good systems outlast good intentions. Clean architecture, modular components, proper testing.
  3. Measure twice — Test the work, then ship it. Playwright E2E testing, CI/CD pipelines, build verification before every deployment.
  4. People over tech — Technology works for humans, not the reverse. Accessible, performant, user-focused design.

Additional principles: "Show, don't tell" (results over claims), "Simple scales" (clarity over complexity), "Test the seams" (systems fail where they connect), "Tools serve people" (technology serves humans).

Template Tier System

Each client site is built on one of four proven tiers. Highland Cleaners demonstrates all four tiers for the same business.

  • Tier 1: Essential — Single-page landing with smooth scroll navigation, contact form, Google Maps, service cards. Ideal for businesses needing a clean web presence.
  • Tier 2: Professional — Multi-page with React Router, 3-step booking wizard, photo gallery with lightbox, form validation. Ideal for service businesses with appointment scheduling.
  • Tier 3: Advanced — E-commerce with shopping cart, product search and real-time filtering, category navigation, analytics dashboard. Ideal for businesses selling products online.
  • Tier 4: Enterprise — Full authentication system, admin dashboard with progress rings and sparkline charts, toast notifications, API integration. Ideal for SaaS or businesses needing user management.

All tiers achieve Lighthouse scores of 95+ Performance, 100 Accessibility, 100 Best Practices, and 100 SEO. All sites meet WCAG AA accessibility standards.

Writing & Articles

Matthew writes about AI-assisted development at projectlavos.com/blog under the heading "Beyond the Code."

The Verification Gap (February 2026)

An exploration of verification challenges in AI-assisted development. When AI generates code that looks syntactically correct, how do humans verify it actually works? The article examines the gap between generation and validation — the difference between "looks right" and "works right." Covers source code inspection vs. runtime behavior, the limits of reading as verification, and practical approaches to closing the gap. Download the PDF.

Guitar Scale Reference Library

24 guitar scale reference books available as downloadable PDFs. Each book covers 12 keys, 5 CAGED positions, 8 exercise types, and 480 total exercises with mathematically verified tablature. All tablature is generated by computational fretboard logic using Python — not hand-transcribed or AI-generated — because LLMs cannot reliably count frets or compute musical intervals.

Scale categories and books:

  • Pentatonic & Blues (4): Minor Pentatonic, Major Pentatonic, Blues, Major Blues
  • Diatonic Modes (7): Natural Major (Ionian), Natural Minor (Aeolian), Dorian, Phrygian, Lydian, Mixolydian, Locrian
  • Harmonic Minor Family (2): Harmonic Minor, Phrygian Dominant
  • Melodic Minor Family (3): Melodic Minor (Jazz Minor), Lydian Dominant, Altered
  • Symmetric Scales (3): Whole Tone, Half-Whole Diminished, Whole-Half Diminished
  • Bebop (1): Bebop Dominant
  • Exotic Scales (4): Double Harmonic (Byzantine/Arabic), Hungarian Minor, Hirajoshi (Japanese), In Sen (Japanese)

Total: 24 books x 480 exercises = 11,520 exercises across all scales. This library connects to the FretVision guitar learning platform.

The Manifesto

"What One Person Can Build" — a manifesto demonstrating what a solo developer with AI collaboration can accomplish. Evidence: 69 websites built for Louisville businesses (live, deployed, working), 85 AI models trained for specific domains (career, code, guitar, analysis), 90+ live URLs all verified HTTP 200, all by 1 person with AI collaboration. Contrasts "The Old Script" (get credentials, ask permission, wait to be chosen, prove you belong) with "The New Script" (build it, ship it, show the work, let the evidence speak).

How Matthew Works With AI

AI collaboration is central to the workflow, but with clear boundaries. AI handles implementation speed — scaffolding components, writing boilerplate, generating test cases, producing variations across 69 sites. Matthew handles all judgment calls — architecture decisions, design quality, business context, and verification that things actually work. LLMs cannot do reliable arithmetic or music theory, so quantitative work (scale calculations, financial formulas, data validation) always uses Python, not prompts. He maintains 85 custom-trained Ollama models for domain-specific tasks and uses Claude for production integrations. This approach is documented in The Verification Gap.

Technology Stack

Frontend: React, Vite, Tailwind CSS v3, Framer Motion, React Router, Three.js (9 distinct 3D formations: Voronoi, L-System Tree, Torus Knot, DNA Helix, Neural Mesh, Orbital System, Lorenz Attractor, Mobius Strip, Governance Cascade), p5.js, Web Audio API, Canvas API, D3.

Backend: Python 3.14, FastAPI, Supabase, PostgreSQL (Neon).

Deployment: Vercel (frontend hosting, edge middleware), Render (backend APIs), GitHub Actions CI/CD.

AI & Machine Learning: 85 custom Ollama models, Claude API, spaCy NLP pipelines.

Testing & Quality: Playwright E2E (328+ test specs), Lighthouse 95+ Performance / 100 Accessibility / 100 Best Practices / 100 SEO across all template tiers, WCAG AA compliance.

Machine-Readable Data & Structured Formats

This site provides multiple machine-readable formats for search engines, AI crawlers, and automated tools:

  • sitemap.xml — XML sitemap for search engine crawlers
  • robots.txt — Crawler directives
  • llms.txt — Concise AI-readable portfolio summary
  • llms-full.txt — Extended AI reference with full portfolio descriptions, methodology, and citable facts
  • portfolio-urls.json — Structured JSON of all portfolio URLs
  • JSON-LD schemas embedded in page source: Person, ProfessionalService, FAQPage (12 questions), ItemList (69 sites + 4 apps), SoftwareApplication (6 apps), CollectionPage, Blog with BlogPosting, BreadcrumbList, HowTo, SiteNavigationElement, WebPage

Site Map