About

Builder of AI-native knowledge infrastructure: systems that make LLMs reliable, auditable, and compounding. Practitioner and methodologist. I ship the systems and document the patterns, including where they break.

Position

Everything here shares one thesis: how to make LLM systems reliable, auditable, and compounding instead of impressive-but-untrustworthy. Provenance on every source, claims you can cite, verification across model families, and an honest audit of the method — including where it breaks.

Reliable AI

Provenance on every source, claims you can cite, verification across model families, and honest failure modes. The hireable signal.

Agent-native architecture

The LLM-wiki pattern: markdown stays canonical, the graph is a query layer. Context engineering, claim registries, subagent-driven development.

Designing for constraint

Reliability and legibility treated as accessibility properties. Honest complexity without shame, applied to engineering.

Method

The proof is the dogfooding: these pieces are drafted from a working knowledge base, with provenance already attached. The system that writes the posts is the thing the posts are about.

Mechanics

Type set in EB Garamond (display) and Source Serif 4 (text), with Inter Tight for chrome and JetBrains Mono for code. Server-rendered with React Router, served from Cloudflare Workers.

Reach

By feed, and by email — sparingly.