IT systems engineer (regulated healthcare + federal) . agent-systems builder
I engineer the systems that keep regulated environments running and defensible -- fleet-scale automation across ~1,400 endpoints, platform migrations led end to end, HIPAA security programs stood up from zero -- and I ship multi-agent AI systems that do real operational work, not demos. US Army veteran with a TBI; most of what I build started as prosthetic executive function for my own brain and grew from there.
- fleet-cve-scanner -- open-source, single-script CVE scanner for RMM-managed fleets. Pure PowerShell 7: joins your RMM software inventory against NVD, CISA KEV, EPSS, and SSVC. The clean-room release of a production HIPAA vuln-management program that surfaced 700+ hidden High/Critical findings on its first run and replaced $15-60K/yr of commercial tooling. Apache-2.0, eight test suites, honest docs about where CPE matching lies to you.
- stellar-notes -- privacy-first voice intelligence across iOS, macOS, and Windows. Shared TypeScript core, on-device ASR (Apple Neural Engine / Parakeet / whisper.cpp), zero backend. Work Mode is a hard per-note privacy boundary. Curated public showcase of the architecture.
- ade -- Windows 11 x64 port of per-simmons/damon-ade, the agent-harness runtime I develop against daily.
- Pythia -- an always-on world monitor:
14 verified live feeds across AI, cyber, geopolitics, US healthcare regulation and
markets, deterministic NEW/CHANGED/GONE change detection with no LLM in that path, and
one cited brief pushed to my phone each morning for $0.008 a day. Every source URL
comes from the evidence store, never from the model. 262 tests.
It started as a multi-model forecasting swarm, and I retired it -- 1,009 forecasts
and 384 resolutions later, a trivial base rate beat it roughly fourfold, so the
experiment is archived on the
review-fixesbranch andmaincarries only the monitor. Began as a fork of jangles-byte/Pythia; credit to the original authors.
Self-hosted AI assistant patterns: run an always-on personal agent on your own hardware -- Telegram front-end, MCP tools, cron, multi-model delegation, and the ops runbook that keeps it alive. Now a full guide + template kit (27 files) with four free MIT starter skills. quietdaemon.com
- LLM-maintained wiki method -- the schema and workflows for a knowledge base an agent maintains and a human curates: typed edges, provenance, lint passes, ingest discipline. Runs my life and ran an IT department (~228 pages across 2 wikis). The third implementation went further and served that corpus to coworkers' agents over MCP -- keyword retrieval, no embeddings and no vector database, 9/9 on an acceptance benchmark recorded before the server existed.
- Securing AI agents in regulated environments -- MCP server security, OWASP LLM Top 10, agent-tooling review, AI governance under HIPAA/BAA.
- kylewelch.net -- professional site + resume
- quietdaemon.com -- the external-brain writing