L'Oréal, Crocs, Lacoste, Breitling, etc. — a small slice of the list.
Always had code under the design work. In March 2026 I went all-in. This was not the plan. Having way too much fun.

5 years designing things for brands you've heard of — always with a scripting layer underneath. Custom Houdini functions and sim setups. Render farms. Project-management app experiments. And one production pipeline that automated rendering 6000 CGI videos through hundreds of Python scripts gluing C4D, After Effects, Houdini, ffmpeg, Blender, and whatever else the deliverable needed.

It always felt like part of the job, not a career. Then in March 2026 I asked Claude to help with a script. Then I asked for one more thing. Then I forgot to stop. Somewhere in there it stopped being a layer under design work and became the thing.

20+ projects on disk now, ~104B Claude tokens through the wire, somehow #1 globally on usage. Some ship to production. Some are research. Some started as one thing and accidentally became another. None were the plan.

• Autonomous research pipelines — orchestrating swarms of headless AI workers across isolated git worktrees. Stream-JSON parsing of subprocess output, quota-wall recovery (parse → sleep → resume), exhaustion detection, supervisor-of-supervisors crash recovery, multi-channel alerting across 6 fan-outs.
• Sub-millisecond reactive systems in Rust — 8-crate workspace, Tokio + lock-free shared-memory IPC, 11 simultaneous venue feeds, deterministic simulation harness with production parity, atomic SQLite-WAL → Parquet snapshots, cross-compilation to ARM production targets.
• Gamified knowledge platforms — React 19 + TypeScript, 800+ questions across 40+ topics, additive schema migrations through 15+ versions, server-side merge-on-write for offline-first multi-device sync, confidence-vs-accuracy bias correction.
• LLM-orchestrated workflow systems — Python + NetworkX DAG analysis with critical-path resolution, dual-priority decomposition (original intent vs current actionable state), conversational interface with hard ASK-PROPOSE-CONFIRM gates.

Four different systems. Won't tell you which is which.

• Multi-agent throughput, not token efficiency — fanning out work across parallel Claude workers to squeeze maximum production speed. Tokens are cheap; calendar time is expensive.
• Custom Claude Code setup — 10+ domain-specific skills, 9+ slash commands, 8+ hooks across projects for safety gates, rule re-injection, doc freshness checks, and operational discipline
• Project memory discipline — 448 markdown files across 20 active projects: feedback, lessons, architecture decisions, session handoffs
• Hard pre-commit gates — AI code critic, doc freshness cascades, schema versioning enforcement, supply chain audits, RustSec
• Operating rules (from my global CLAUDE.md) — no silent decisions, verify every fix with real data, surgical changes only, root causes not workarounds, no auto-pause

Rust (Tokio · Alloy · ONNX · iceoryx2 · sqlx) · Python (Flask · NetworkX · httpx · pytest) · TypeScript / React 19 / Vite / Zustand / TanStack Query · whatever else the problem needs

• Designer for 5 years · developer since March 2026
• ~104B Claude tokens · #1 globally on ccgather
• ~1.7B tokens / day average · 1,828 sessions
• 20+ active projects · 448 memory files
• In one project alone: 50 architecture decision records, 12 zero-bypass pre-commit gates, 8 Rust crates, 11 venue feeds
• In another: 820 exam questions, 41 topics, 15 schema versions