Software developer working across application, backend, and infrastructure layers. Currently focused on self-hosted systems and local model serving.
Clarks Summit, PA · Member of Susquehanna Syntax
DesignRNG — desktop authoring tool
An Electron desktop application for authoring procedurally-generated content packs. The 3D editor is built on React Three Fiber — transform gizmos with legal-angle rotation, face-resize handles, and a snap-point editor that defines where each part docks on a chassis. A Python sidecar speaks JSON-RPC to the renderer and owns schema validation and packaged export, which keeps the domain logic out of the UI process.
Decisions are recorded as 8 architecture decision records, with supersession tracked when a decision replaces an earlier one.
Arms NG — JVM runtime
The runtime that consumes those packs — a Minecraft Forge mod that composes weapons from authored parts at load time. It ships no built-in content: everything is resolved from the pack on disk.
These are two separate programs, in two languages, maintained in two repositories. What connects them is a versioned wire schema with its own fixtures and a drift check that runs in CI, so a change to the authoring side cannot silently break the runtime. Designing and holding that contract is the work I'd point to first.
| Languages | Python · Java · TypeScript · JavaScript · GLSL · Bash / PowerShell |
| Application | Electron · React · React Three Fiber · three.js · Zustand · Vite · PySide6 / Qt |
| Backend | ASGI / Uvicorn · httpx · JSON-RPC · SQLite · REST |
| JVM | Java 25 · Gradle (Groovy + Kotlin DSL) |
| Infrastructure | Linux · Raspberry Pi OS · systemd · image builds · GitHub Actions |
| Testing | pytest · Vitest · fixture + schema-drift verification |
Self-hosted model serving. A provider-agnostic broker exposing one Anthropic-Messages-compatible API, routing requests to hosted APIs or local CLI agents, with preflight guardrails, run history, and a small companion UI. Target deployment is a Raspberry Pi Zero 2 W running Pi OS Lite. Status: working skeleton — not yet deployed.
GPU hardware evaluation. Assessing RTX 3090-class hardware as the basis for a local inference server, starting with a single GPU before deciding whether to expand. Status: evaluating — nothing purchased, benchmarked, or deployed.
- Architecture decision records — decisions written down and locked, supersession tracked
- A domain glossary per project — one agreed word per concept, so naming stays stable
- Versioned wire contracts — cross-language boundaries specified, fixture-tested, drift-checked
- CI and release automation — GitHub Actions for tests, schema checks, and packaged builds
Some of my work is in private repositories.