writing 8 public repos

case studies

Writing

Neutral studies of systems I've built and run — what each is, and what shipped.

Spaces — a desktop platform that freezes and thaws whole workspaces

Organizes a machine into Spaces — each with its own apps, layout, and local-model context — that freeze and thaw on a keystroke. GNOME + Hyprland from a shared core.

platformssolutions read ▸

mapit → shipit — a five-stage delivery pipeline

A five-stage promotion pipeline (T0 → T1 → T2) implemented as tooling, plus the continuity and decision artifacts that make engagements repeatable.

solutionsdelivery read ▸

System cartography — schema-first maps and typed graphs

A wide-net cartography method that catalogs an unfamiliar domain before any build, and two typed graphs that make a corpus queryable.

retrievalmapping read ▸

Enforcement RCA — a root-cause program for a live ML model

A repeatable root-cause-analysis program for a live ML enforcement system. Methodology only — no platform-internal data.

trust-safetygovernance read ▸

Governance patterns — instruments that keep a rule enforced

An adversarial multi-judge review panel and an access control paired with a reconciliation loop — two shapes for turning a declared rule into an enforced one.

governancesafety read ▸

gfx1150-bench — inference tuning on a handheld iGPU

A controlled performance study on an AMD handheld iGPU: roofline, an 8-model ROCm-vs-Vulkan matrix, and a 2.43× speculative-decoding result.

gfx1150-bench ai-infrainference read ▸

Industrial edge — a sprayer diagnostic, a field curriculum, a protocol library

A CAN/GPS diagnostic on agricultural spray equipment, a field curriculum, and the public protocol library that backs the domain.

opendbc-ag industrialedge read ▸

SecondChair — self-hosted legal RAG

A self-hosted, privilege-aware legal retrieval-and-analysis system built with a small practice; v1 shipped to production in a seven-day engagement.

retrievalsolutions read ▸

Reproducible, keyless agent evals

A record/replay seam that lets an agent-eval harness reproduce its scorecard in CI with no API key — plus the deterministic hard-gate mechanism that rides on top of it.

agentic-eval-harness agent-infrareproducibility read ▸

inkSpec — one engine, any design system

A no-build, offline, single-file HTML engine that renders LLM-authorable specs as forms and dashboards — and ingests any third-party design system as one CSS file of tokens against a thin, enforced contract.

spec-renderer platformsagent-infra read ▸