Selected Work
Recruiter-relevant projects across agent tooling, systems performance, ML/detection, and privacy-first infrastructure. I keep this list intentionally curated.
Recent Upstream Impact
"Ghost in the Codex Machine" Fix
Root-caused and fixed a pre-main environment regression that stripped LD_* / DYLD_* env vars, triggering CUDA/MKL fallbacks and large slowdowns for some users.
Agent Tooling (LLM-Centric)
llmx
What: Local-first codebase indexer with BM25 search and semantic chunk exports (agents can retrieve only what they need).
Why it matters: Retrieval that is fast, deterministic, and privacy-preserving (no embedding service required).
codex-xtreme (includes codex-patcher)
What: An interactive wizard for producing optimized, patched Codex binaries, backed by a verified patch application engine.
Why it matters: Codifies the "agent edit loop" into explicit, debuggable steps.
Primary repo (pin this on GitHub): codex-xtreme
Patch engine (linked inside): codex-patcher
burn-plugin
What: Claude Code plugin for the Burn deep learning framework (skills + workflows + evidence-driven references).
Why it matters: Demonstrates how to package domain expertise into an agent workflow that can be reused.
cwork
What: A context compiler for Claude Code workflows that composes base capabilities + domain primers + project instructions into a minimal, task-specific context.
Why it matters: Skill systems and context management are where multi-repo agent workflows succeed or fail.
claude-cowork-linux
What: Run the official Claude Desktop app on Linux using compatibility stubs and a bubblewrap sandbox.
Why it matters: Practical agent workflow enablement (Linux-first, security-aware).
Note: Unofficial community project; no proprietary Claude code is committed.
Iconics
What: Semantic icon library (8k+ icons) designed to replace emojis with consistent PNG icons.
Why it matters: Professional documentation UX at scale (and it's agent-friendly: meaning-based search, deterministic exports).
MCP Servers (Tool Surfaces for Agents)
pyghidra-lite
What: Token-efficient MCP server exposing a structured tool API for program analysis workflows (compact output by default, opt-in verbosity).
Registry: Official MCP registry: io.github.johnzfitch/pyghidra-lite (v0.1.1, status: active, published 2026-01-29).
Why it matters: In practice, reliable agents are tool-driven. MCP servers are how you turn a complex system into a safe, inspectable interface.
Systems Performance (Real Data Structures, Real Constraints)
Triglyph / Triglyphd
What: Zero-RSS trigram index with custom binary formats + a D-Bus daemon for system-wide search.
Why it matters: Low-level performance engineering (mmap, layout, predictable latency).
filearchy
What: COSMIC Files fork with io_uring backend and trigram search integration.
Why it matters: Applies systems work in a real product surface (UX + async I/O + search).
ML / Detection (Algorithm-First, Production-Aware)
Observatory
What: AI image detection suite running 4 models entirely client-side (WebGPU/WASM).
Why it matters: Real product constraints: model size, caching, orchestration, UX, and "no server required."
Live: look.definitelynot.ai
SpecHO v2
What: 161-dimensional linguistic fingerprinting system for AI text detection and model identification (tiered feature pipeline).
Why it matters: Algorithm-first features, measured discriminators, and clear engineering of runtime tiers.
Security and Privacy (Defensive)
definitelynot.ai
What: Unicode-security-aware text sanitizer with Trojan Source defense, homoglyph mitigation, and BiDi neutralization.
Why it matters: Practical security tooling with a clear UX and an API for integration.
Infrastructure (Self-Hosted)
digitaldelusion
What: Bare-metal NixOS infrastructure with authoritative DNS, automated wildcard certs (DNS-01 / RFC2136), and post-quantum VPN (WireGuard + Rosenpass).
Why it matters: Demonstrates production ops maturity, reproducibility, and security-first design.
Writing / Research (Available on Request)
AURORA Protocol
What: Design work on robust agent communication in hostile networks (message framing, discovery, encryption-by-default, federation).
Why it matters: Systems thinking across protocols, security, and agent interoperability.
Selected Private Work (Names Only)
- eero (private)
- alienware-monitor (private)
- proxyforge (private)