AI Architect / Builder / Maker

Dominick
DeFazio

Building production LLM systems for enterprises across regulated industries. Multi-agent architectures, generative UI, and the infrastructure that makes AI actually work.

View Projects Skills AI Art LinkedIn GitHub
Selected Work

Projects

Systems I've designed and built — from multi-agent pipelines orchestrating 15+ specialized AI agents, to things I make for fun on a Thursday night.

Agent Architecture
Multi-Agent System

Agentic Infrastructure

Full documentation of a multi-agent build pipeline — 5 teams, 15+ specialized agents (Architect, Mason, Artisan, Gatekeeper, Scribe...), orchestrated by a CEO agent with mandatory quality gates, TDD enforcement, and session recovery. Built on Claude Code.

Explore the architecture
Tarot Game
Interactive / One-Hour Build

Tarot

An interactive tarot card experience with physics-based card mechanics, starfield backgrounds, and fan-out reveals. Built in a single hour as a creative exploration — 876 lines of vanilla JS, zero dependencies.

Draw a card
vitalChess Leaderboard
Live PWA

vitalChess

A local-first PWA tracking chess club matches and ELO rankings. Top-3 leaderboard treatment, animated ELO deltas, PIN-based player auth, Supabase cloud sync with realtime viewer mode. 1366 tests across 119 test files. Built end-to-end by the 18-agent pipeline.

Visit vitalchess.com
Custom Tooling

Skills

Custom skills built into Claude Code that orchestrate agent teams, scaffold entire projects, run autonomous deliberation, and analyze decision history. The infrastructure that makes work compound.

4-Agent System

atelier

Multi-agent programmatic art studio. Spawns four specialized agents — Scout, Composer, Renderer, Critic — that research, plan, generate, and critique visual artwork rendered as code (node-canvas, GLSL shaders, geometric layouts, pixel art). Produced the gallery pieces below.

See how it works
18-Agent Pipeline

init-pipeline

Initializes a new project with the 18-agent, 5-team build pipeline. Creates .claude/ agents, skills, rules, and project scaffolds from templates. The system that builds the systems.

See how it works
Decision Analytics

analyze-decisions

Analyzes a SQLite log of AskUserQuestion answers across all projects — surfacing preference signals, frame rejections, contradictions, and memory candidates. Tooling that mines a user's decision history to improve agent alignment over time.

See how it works
12-Delegate Council

quick-assembly

Spawns a Model UN-style General Assembly of 12 thinkers to deliberate on a single hard question. Delegates research, form organic blocs, draft competing resolutions, negotiate amendments, and vote on a binding outcome — fully autonomously after the brief.

See how it works