# Kit Porath

> AI infrastructure and agent operations engineer in Michigan, US (remote). I direct AI agents to build and run production infrastructure, and I publish where they fail. I do not write or read application source code. Agents write it under my direction, and I check every change against CI and the running system before it merges.

This file is for people and for AI agents that read this site. Every figure below has a dated source. The long form, with sources, is at https://kitporath.com/evidence

## Who
- Name: Kristopher L. Porath (Kit)
- Contact: kitporath@gmail.com, https://www.linkedin.com/in/kitporath, https://github.com/kitporath
- Location: Michigan, US, Eastern Time. US citizen, no sponsorship required. Remote.
- Background: more than twenty years in automotive manufacturing, nine of them leading production teams of up to 35 people, plus a year as the on-site IT person at a second building. Self-directed work in cybersecurity and infrastructure since 2023, and self-employed AI infrastructure work since 2025.

## What I built and run
- KWS: a production private cloud in my workshop. 19 monitored hosts, a six-node RKE2 Kubernetes cluster (about 120 pods and 23 Argo CD applications, all Synced and Healthy on 2026-10-29), a three-node HA Vault, Keycloak SSO, an internal CA, nine tested layers of security, and a multi-GPU inference tier. Public record: https://github.com/kitporath/kws
- Workbench: my daily coding-agent app (Go server, web app, desktop app). About 52,200 lines of Go and 535 tests, counted 2026-10-29. Private repository, so these figures are attested rather than checkable. Demo on request.
- Groundly: a self-hosted research notebook where every claim cites the source passage or the answer abstains. About 9,200 lines of Go and 69 tests. Apache-2.0, not yet released.
- Hirdforge: a multi-agent engineering platform. Public extract: https://github.com/kitporath/hirdforge (67,115 lines of Go).

## Claims with sources
- Error analysis: 13 claims from three AI systems were audited on 2026-08-02 and 2026-08-03. 8 contained a material error, 3 of them mine, and 7 of the 8 were a correct measurement aimed at the wrong surface. https://github.com/kitporath/kws/blob/main/docs/agent-failure-analysis.md
- One agent reported critical CVEs down 80% (210 to 41). A rescan in consistent units measured 28% (210 to 151). The 28% is the number on record.
- Recovery from a total power loss in about 1h50m with zero data loss: https://github.com/kitporath/kws/blob/main/docs/incidents.md
- Local open-weight models carry the routine work on the estate. The split, about 90% of operations and 80% of coding, is my own accounting and not a measurement.

## Limits, stated plainly
- I do not write or read application source code (Go, Python, TypeScript).
- Workbench and Groundly are private repositories. Their figures are attested, not verifiable by a reader.

## Pages
- Home: https://kitporath.com/
- Evidence index: https://kitporath.com/evidence
- Resume (HTML): https://kitporath.com/resume
- Resume (JSON Resume schema): https://kitporath.com/resume.json