Senior Systems Engineer · MSP delivery · 10 years in IT (USAF) · Colorado
Linux desktop, infrastructure visibility, human-gated AI development, local agent compute
I design and deliver infrastructure for more than 80 client environments by day. Off the clock I build the systems I run on: a desktop that understands the machine under it, a control plane that keeps AI agents on a leash, a live view of my lab, and an agent stack that runs on my own GPU. These are not four side projects. They are four layers of one environment.
| Layer | System | Status |
|---|---|---|
| BUILD | Aphotic-Hypr | public, beta |
| BUILD | GATE | public, early |
| OPERATE | AETHER | private, live lab |
| COMPUTE | Claude-Local | local architecture |
The four projects people actually use. Star counts update on their own.
| Project | What it is | |
|---|---|---|
| 1 | berth ·  |
Decides who gets your GPU. Chat, image, and voice engines lease the card in turn instead of grabbing it, so two of them never load at once and crash each other. One Go binary. |
| 2 | aphotic-hypr ·  |
Four identities, one desktop. Modular plugins, live hot-swapping, resource-aware. |
| 3 | oh-my-pocket ·  |
Pi Pocket tuned for local models, running on your machine or on your phone. fork |
| 4 | oh-my-epiq ·  |
Local-first, git-backed issue tracker with a terminal, a browser board, and an agent MCP. |
Everything else pushed in the last 60 days, grouped by what it does. These are the fun experiments I am building now. 🔒 marks a private repo, which has no public link.
- aphotic-plugins ships the official plugins for Aphotic-Hypr.
public - aphotic-themes ships the official theme pack for Aphotic-Hypr.
public - aphotic-pets lists themable pets for Aphotic-Hypr.
public
- GATE is a localhost control plane where a human decides every gate.
public, early - 🔒 aether-harness is a local-first agent harness that keeps its log through a crash or a dropped terminal.
private, early - ModelHelm delegates work across local and cloud models over MCP.
public - llm-tune is an agent skill that tunes local LLM settings against measured evidence.
public
- 🔒 homelab keeps the inventory, runbooks, and decisions that operate the lab, with a public journey log in
public/.private - 🔒 aether-grid is a live dashboard for local inference, fed by Speculum.
private - 🔒 ai-blog is the AETHER public site, private until launch.
private - 🔒 BlastRadius maps which services and devices a proposed infrastructure change affects.
private, concept - 🔒 InfraSuite is a design for AI-assisted infrastructure operations where an enforcement layer, not the model, decides what may run.
private, concept - 🔒 M365-Tools collects PowerShell tools for Microsoft 365 administration, migration, and security hygiene.
private
- 🔒 berth-private drains the card on handoff and reports VRAM squatters.
private - speculum watches a local LLM runtime and draws it on one page.
public - 🔒 spectra explains why a GPU workload performs the way it does, with the evidence attached.
private - 🔒 AdaScope ranks the likely bottlenecks in GPU inference and proposes the smallest experiment to confirm them.
private - 🔒 SMTrace fingerprints CUDA workloads and proposes optimization experiments.
private - ninfer-4090 runs Qwen3.8-27B on one RTX 4090 with a full 262K context.
public, fork - strata installs Qwen3.8-Flash-Next on consumer hardware in one step.
public, fork - llama-swap swaps models behind any local OpenAI or Anthropic compatible server.
public, fork - local-ai-registry stores one validated recipe per machine, with the evidence attached.
public, fork - llama.cpp is the GGUF inference runtime behind the rest.
public, fork - ComfyUI-QwenVL-Mod adds Qwen-VL with GGUF support to ComfyUI.
public, fork
- Strix-Recon automates reconnaissance for penetration tests with an LLM.
public
Archive, older and inactive projects
- worldsimulator renders a browser geospatial intelligence prototype.
public, 2026-02 - blackarch tracks the Arch-based distribution for penetration testers.
public, fork - Meraki-EOL-Manager manages Meraki hardware lifecycles.
public - hyprdots collects dotfiles for Arch and Hyprland.
public, fork - secrets collects ffuf and fuzzing notes.
public - hyprland-layoutswitch switches keyboard layouts under Hyprland.
public - Hacktoberfest2023 holds a Hacktoberfest pull request.
public, fork - github-readme-stats generates README stats.
public, fork
Inference stack, workstation, and tooling
| Layer | Tool | Role |
|---|---|---|
| Orchestration | Claude Code | agent harness, subagents, tool execution |
| Local and cloud split | claude-router | local roles to llama-swap, cloud traffic passed through |
| Routing | llama-swap | role aliases, swap groups, load and unload lifecycle |
| Runtime | llama.cpp and llama-server | GGUF inference, pinned builds, slot, context, and reasoning-budget config |
| Serving (alternative) | Ollama, LM Studio | quick model management, desktop serving |
| Training | Unsloth, PyTorch | fine-tuning experiments |
| Compute | RTX 4090, CUDA | 24GB VRAM |
Working knowledge. GGUF and quantized inference (IQ and Q4 quants, MoE versus dense), local model serving, GPU inference and VRAM budgeting, model routing, runtime configuration for slots and context, context management, agent orchestration, multi-agent workflows, local coding agents, local and cloud interoperability over OpenAI-compatible APIs, and task-level evals that pick which model earns which role.
Workstation. An Intel Core i9-14900K, an RTX 4090 with 24GB VRAM, 64GB RAM, and a custom water loop. It boots Arch Linux for daily use, Aphotic development, and the Claude-Local host, plus Windows 11 when the job needs it. Both boots share the same llama-swap stack.
Infrastructure and enterprise stack
Security, Hack The Box
Hack The Box as 5H3LLKiller. Kali and BlackArch tooling, plus the Aphotic Security profile for offensive-research sublayers. Hack → Scream → Repeat.
Ask me about Linux, Proxmox, MSP infrastructure, Hack The Box, or local inference. Help on Aphotic-Hypr is always welcome.