I build and ship software with Claude Code and Codex as my daily workflow. I set the direction, reproduce the problem, review every diff, and run the tests; the agents do the typing. Everything below is public and linked so you can check it.
What I'm good at hiring-wise: turning a vague bug report or feature ask into a reproduced, tested, merged change inside someone else's codebase; building small end-to-end products (desktop, web, Android); and making agent workflows cheap and reliable enough to run all day.
| Merged upstream pull requests | 48 across HiveMind (46), Protocol-Canary #570 and Credence-Contracts #1636 |
| Under review right now | 6 PRs in excalidraw, highlight.js and mermaid |
| Shipped products | PC Cast & TV apps (4 desktop editions), Hatchpile (web + Android), Lucent and Synapse (browser GPU demos) |
| Languages in the work | TypeScript, Python, Rust, Java, PowerShell, GDScript, Lean 4, WebGL2 |
| Tooling I built for agent work | claude-code-toolkit: hooks, local-model delegation, cost measurement |
Casting, live-TV guides and remote control for PCs and TVs. Four Windows/Linux desktop editions, Android betas and a mobile Live web app. Python, packaged and sold as IN3RT1A.
Releases · Mobile Live |
A cozy physics merge puzzle with a shared daily puzzle. One codebase runs as an installable web app and as a signed Android APK via Capacitor.
Play · Android |
A progressive GPU path tracer in TypeScript and WebGL2: glTF import, physically based materials, HDR lighting, and a render budget you pick for your GPU.
Live demo · Engineering notes |
A browser trace explorer: service map, waterfalls and a live globe of Wikimedia edits. Imports OTLP/JSON, Jaeger and Zipkin traces.
Live demo · Architecture |
| Project | What I did | Status |
|---|---|---|
| HiveMind | QUIC 0-RTT and STUN/TURN networking with loopback tests (#40, #73), INT4/FP8 fused router and CUDA-stream prefetch kernels (#23, #7), error-feedback compression (#63, #69), Lean 4 conservation proofs (#68), CI and benchmark repairs (#67, #77) | 46 merged |
| Credence-Contracts | Soroban bounty-escrow entry points and paginated views in Rust | #1636 merged |
| Protocol-Canary | getHealth RPC compatibility |
#570 merged |
| excalidraw | Arrow-type switching for bound and labeled arrows, stable canvas search results while editing, searchable shortcuts in the help dialog | #12503 · #12504 · #12505 open |
| highlight.js | PHP one-line comments now end at the closing ?>; Ruby Name::method no longer read as a symbol |
#4572 · #4573 open |
| mermaid | Build plugin that strips chevrotain's unreachable eval from the minified bundles (fixes a Firefox console warning) |
#8649 open |
Every one of these started the same way: reproduce the report against the current source, write the test that fails, make it pass, run the project's own lint and suite, then open one focused PR.
- Reproduce first. No fix goes out for a bug I could not trigger locally. If a report does not reproduce, I say so instead of guessing.
- Tests that fail before and pass after. Both runs go in the PR description.
- Cheap, long sessions. I wrote claude-code-toolkit: hooks that block repeat file reads, retire long-running agents before context cost compounds, shrink screenshots before a model sees them, and a wrapper that sends mechanical work to a free local model, with a ledger of what that saved.
- Bounded claims. My research repos (Godot rollback coherence, Ruffle portable states, Adobe Tone Lab) state exactly what was tested and what was not, including the experiments that failed.
Completed the requirements for an Associate of Applied Science in Networking & IT Security at Muskegon Community College. Self-taught in everything above since, mostly by shipping.
Stack I reach for: TypeScript · Python · Rust · WebGL2 · Lean 4 · Linux · Godot · QUIC/STUN/TURN · Claude Code · Codex
Open to contract and freelance work: shipping features in your codebase with agent-assisted workflows, integrations, debugging, build and release tooling, and setting up Claude Code or Codex so your team gets the same results.
Open an issue to start a conversation · Full project catalog · Apps