Learning Materials: AI Infrastructure, Networking & Purpose

Over the past few weeks, I've been building out my personal AI infrastructure — a multi-agent system spanning a MacBook Pro, Mac Mini M4, Raspberry Pi 5, and a Linux desktop, all connected via Tailscale. Along the way, I've created a series of interactive slidedecks that document the architecture decisions, hard-won lessons, and practical guides that came out of this work.

Each slidedeck below is a self-contained, interactive HTML presentation that runs entirely in your browser — no installs, no dependencies, no tracking. Just download the HTML file, open it, and use arrow keys to navigate. They're designed with a dark-mode aesthetic and monospace typography, optimized for engineers who prefer substance over style.

Whether you're setting up a headless Mac server, exploring multi-agent AI architectures, running LLMs on a Raspberry Pi, building a Tailscale monitoring stack, or searching for your life's purpose, there's something here for you.

Mac Mini M4 Headless Server Guide

Running a Mac Mini M4 as a headless server sounds straightforward until you actually try it. Apple Silicon introduces a constellation of quirks that make headless operation surprisingly tricky — from aggressive sleep behavior that ignores your Energy Saver settings, to screen resolution defaulting to 1024x768 without a display, to FileVault encryption preventing remote boot without a connected keyboard.

This 10-slide guide walks through every gotcha I encountered turning a Mac Mini M4 into a reliable 24/7 server. It covers SSH setup, the multi-layered approach needed to actually prevent sleep (hint: you need caffeinate, pmset, AND an Amphetamine-style approach), HDMI dummy plugs for proper resolution, VNC/Screen Sharing configuration, and the critical FileVault + auto-login dance. There's a full command cheatsheet and a step-by-step setup checklist you can follow with a display attached before going headless.

If you've ever had a Mac Mini go to sleep at 3am and refuse to wake up over SSH, this guide is for you.

Download: Mac Mini M4 Headless Server Guide (10 slides)

Paisley + Ocasia: Dual-Agent Architecture

What happens when you split your AI assistant into two specialized agents — one that lives on your laptop as a deeply-integrated development partner, and another that runs 24/7 on a headless server handling autonomous tasks? This 14-slide deck documents the architecture of my "Partner + Employee" dual-agent system built on Claude Code and OpenClaw.

Paisley (the Partner) runs on my MacBook Pro with full access to my development environment, codebase, and personal context. Ocasia (the Employee) runs on the Mac Mini with web access, Telegram integration, and the ability to operate independently around the clock. The presentation covers the three-layer memory sharing architecture (session, project, shared), the delegation protocol that lets Paisley dispatch tasks to Ocasia, the security model that prevents cross-agent privilege escalation, and a real ROI analysis showing cost-per-interaction.

It also digs into why Telegram became the communication backbone, how the skill systems compare between platforms, and the anti-patterns we learned to avoid (like giving agents too much autonomy too early). If you're thinking about building a multi-agent system for personal or professional use, this is the architecture brief.

Download: Dual-Agent Architecture (14 slides)

AI Agent on Raspberry Pi 5

Can you run a meaningful AI agent on an 8GB ARM board? This 20-slide deep dive — produced by four parallel research agents — answers that question comprehensively. We evaluated 12 agent frameworks (from 5MB Rust binaries to full platforms), benchmarked local LLMs on actual Pi 5 hardware, compared inference engines, tested operating systems, and even looked at alternative single-board computers.

The short answer is yes, but with very specific choices. The presentation identifies the optimal stack: a lightweight Rust-based agent framework (ZeroClaw or Moltis), Ollama as the inference engine, 1-3B parameter models in Q4_K_M quantization, and Ubuntu Server as the OS with specific kernel tuning for LLM workloads. You'll find real benchmark data — tokens per second, memory usage, load times — not theoretical estimates.

The guide also covers hardware essentials (NVMe storage is non-negotiable, active cooling is critical), the Hailo AI Kit verdict (skip it for LLMs), and a gallery of community projects running AI agents on Pi hardware. Whether you're building a home automation brain, a private AI assistant, or just want to understand what's possible on the edge, this is the reference.

Download: AI Agent on Raspberry Pi 5 (20 slides)

Tailscale Monitoring Stack

When you have three hosts on a Tailscale mesh — a MacBook, a Mac Mini, and a Raspberry Pi — running various AI agents and services, you need unified observability. This 18-slide research briefing documents the complete monitoring architecture from agent selection through deployment.

The centerpiece is Grafana Alloy, the successor to Promtail, Grafana Agent, and several other tools that Grafana Labs has consolidated into a single binary. The presentation makes the case for Alloy with real resource footprint comparisons on Pi 5 hardware, complete configuration examples for remote hosts, and a clear explanation of why the hybrid push/pull architecture makes sense for a Tailscale mesh. It covers Tailscale's native metrics (built-in since v1.78), fleet-level visibility via the Tailscale API exporter, and the security model where Tailscale itself provides the encryption and access control layer.

You'll find step-by-step deployment instructions for Raspberry Pi, DigitalOcean Docker hosts, and Mac Mini, along with Prometheus configuration updates, Grafana dashboard IDs to import, and a complete implementation checklist. If you run a homelab or small infrastructure on Tailscale, this is your monitoring playbook.

Download: Tailscale Monitoring Stack (18 slides)

The TELOS Method

This one is different. While the other presentations are deeply technical, The TELOS Method is about something more fundamental: finding your purpose. Based on Dan Miessler's framework (from his work on Unsupervised Learning), this 7-slide explainer presents a structured approach to the question most people struggle with their entire lives — "What should I do with my time here?"

TELOS (from the Greek word for "purpose" or "end goal") provides a practical framework for self-discovery that doesn't rely on vague inspiration or personality tests. The presentation walks through the core problem (most people optimize for proxies like money and status rather than actual fulfillment), introduces the framework's key components, and provides actionable steps for beginning your own exploration.

I included this alongside the technical content because building personal AI infrastructure is, at its core, about amplifying what matters to you. If you haven't figured out what that is yet, this framework is a good place to start.

Download: The TELOS Method (7 slides)

All slidedecks are self-contained HTML files with no external dependencies. Open in any modern browser, navigate with arrow keys. Built with love, caffeine, and entirely too many late nights.