Building and DocumentingAI Agent Systems
A public library of agent engineering, agent infrastructure, and live AI systems. Code and reasoning in the open. Honest about what works and what does not.
Browse by topic.
Each hub lists every article on one topic, in reading order.
Local AI
Open models on two gaming laptops: setup, memory limits, benchmarks, remote access, a private voice assistant, and n8n automations.
AI Security
Bunker, the offline crypto address scanner, plus website hardening, a Zero Trust home lab, and AI safety builds.
AI Governance
An AI acceptable-use policy, an AI tool register, the reasons behind their design, governance builds, and course notes.
Five bodies of work.
Some shipped, some in active development, all real. Taking ideas to working, documented systems, and the reasoning behind them.
Bunker
A free scanner that finds which crypto addresses expose their public key, with a step-by-step plan to hide what can be hidden. It runs only on your device, covers more than 100 networks, and works in English and Spanish, for people and AI agents.
Korvin
A self-hosted AI agent for low-cost servers. Voice-first, memory-persistent, model-agnostic. Telegram interface and web dashboard. No cloud dependency, no lock-in. In active development — not yet released for public download.
Agent Patterns & Builds
A growing public library of documented agent builds across tool use, memory, retrieval, multi-agent orchestration, evaluation, safety, and governance. Each one is a working pattern with the code and the reasoning behind it.
Live AI Systems Series
Most AI is reactive prompt and response. Live AI systems watch changing context, hold state, detect meaningful change, score priority, produce evidence, and escalate to a human when review is needed. A series on building exactly that.
Build notes and teardowns.
50+ articles, organized by category. Architecture teardowns, debugging notes, and operating lessons from real builds. The errors left in.

The Flipped Interaction Pattern: Agent Asks First
Why a good agent interrogates before it acts, with the code and the failure modes.

Three hidden patterns that inflate your AI API bill
Where cost actually compounds in agent workflows, and how to catch it early.

TaskWeaver Stateful Data Analyst
A LiteLLM rebuild of TaskWeaver showing planner-to-executor handoff, stateful data memory, and safe pandas actions.
Building.
I build working AI systems with frontier models and document every step. Each build comes with the code and the reasoning, so others can learn from it.
Background in technology and intellectual property law. That background shapes how I write about AI risk, ownership, and governance.
Every build is open: the code, the architecture, and the reasoning behind each decision. Study it, run it, build on it.
Working on something similar?
Every build here started as a real problem. If your team has one like it, tell me about it. I read every message and reply myself.
Phone: +1 786 247 3470