Engineering in practice. Questions that stay with me. A record of what I’m building, learning, and trying to understand.
New writing on AI safety, security, and the systems we live with.
One Node process, five vaults, a local vector index, a persona agent with a write protocol, and an issue tracker that AI agents share. Seven diagrams of a Personal OS, and what keeping it local actually costs.
From dial-up to data breaches, from "move fast and break things" to philosophy PhDs in AI labs. How security taught us boundaries, privacy taught us that data is a liability, and philosophy is about to teach us the hardest lesson of all.
A control plane for AI coding agents. Why governing agents through prose in CLAUDE.md does not scale, and what a hooks-based enforcement layer with default-deny egress and signed, version-controlled policy looks like in practice.
An amateur experiment with context windows, memory, and what comes out the other end. Two short files loaded into a model's context each turn, and what came back when I asked it about consciousness, determinism, and itself.
A typed identity layer for AI agents. Schemas, evidence-backed claims and bidirectional links across 173 vault files turn freeform user context into a structured API for LLMs.
Twenty years of handwritten diaries as the memory layer of a personal AI. How vault, RAG, OCR and an agent loop come together into a system that knows who you actually are.
Most public MCP demos that read patient data would fail a HIPAA audit on a single grep. A walk-through of a healthcare-grade MCP server design where the patient ID never enters logs by construction.
What changes when a personal knowledge base is about you, decisions, energy, health, inner life. Domain policy files, behavioral contracts for AI agents, and why one librarian with one behavior breaks down.
A structured weekly review agent with MCP read-write access to an Obsidian vault. From capability to application, and why the specificity of the workflow turned out to be the point.
Building a Cyrillic handwriting OCR pipeline for 20+ years of personal diaries. TrOCR, LoRA fine-tuning, and why the hardest part is not the model.
How I built a local REST API over an Obsidian vault, connected it to a React dashboard, and added an MCP layer for AI-driven read-write access. Treating personal knowledge management as infrastructure.
A look behind the curtain of AWS certification creation. How Subject Matter Experts develop and refine exam questions, the workshop format, and what the program is actually like from the inside.