🎓 Academic

• BBA, MBA, MSBA — University of Iowa
• Master of Data Analytics Leadership — Drake (Student of the Year)
• Adjunct since 2018: R, Python, Analytics, Capstone
• You may have had me before 👋

💼 Professional

• CX Analytics Manager @ Agilent Technologies
• Meeker Technologies (consulting LLC, 2026)
• Focus: AI systems that empower, not replace

🏠 Home Base

• Tiffin, Iowa — wife Tiffany, daughters Ellie & Tessa
• Mini bernedoodle named Maui
• Running AI infrastructure from my basement

🤖 PAI — Personal AI Infrastructure

• Orchestrate multiple agents around your goals
• Persistent memory that grows with you
• Workflow automation + self-improvement loops
• 243+ Fabric patterns for content analysis
• Goal: AI that works for you, not the cloud vendor

🌐 Human 3.0 — The Destination

• Human 1.0 — Hunter-gatherer, physical survival
• Human 2.0 — Industrial worker, economic output
• Human 3.0 — AI-augmented, self-directed, creative

🌀 Outer Loop — Your Life

• Where are you now? (current state)
• Where do you want to be? (desired state)
• PAI continuously closes that gap

⚙️ Inner Loop — Daily Operations

• Ingest → Analyze → Synthesize → Act
• Email, calendar, tasks, research, communications
• Runs autonomously — briefings, triage, updates
• Reports back for human decision on exceptions

⚠️ Problems with "one big brain"

• Context windows fill up fast
• No specialization = mediocre at everything
• No memory between sessions
• No way to parallelize work
• No audit trail
• Single point of failure
• Can't run autonomously

What it does

• 🔀 Session management — maintains context per agent
• 🛣️ Agent routing — Barack gets general queries, Vega gets market questions
• ⏰ Cron scheduling — briefings, email checks, automations
• 📡 Multi-channel — Signal, iMessage, Email, web dashboard
• 🔊 TTS integration — voice output on any channel
• 🧠 Memory injection — feeds context to agents

Channels in use

• 📱 Signal — primary messaging (me → system)
• 💬 iMessage — family automations
• 📧 Email — Inbox agent monitors
• 🖥️ Mission Control — web dashboard

Example cron jobs

• 6:45 AM — Morning briefing (Barack)
• Every 30 min — Email triage (Inbox)
• Friday 4 PM — Weekly status report
• Sunday 8 PM — Week preview

🔊 Chatterbox TTS

• Local inference — nothing leaves the house
• MPS-accelerated on Apple M4
• Running at localhost:4126
• 6 cloned voices
• ElevenLabs-compatible API

🎭 Cloned Voices

• 🎩 Obama — Barack, Maxwell (authority)
• 🎸 Taylor Swift — Sage, Taylor (approachable)
• ⚖️ Olivia Benson — Olivia (direct)
• 🎬 David Attenborough — Matlock (gravitas)
• 🎙️ Morgan Freeman — Marcus, Forge, Vega

🔄 Voice Pipeline

mic input
  → Web Speech API (STT)
  → OpenClaw (routing)
  → Agent (Claude)
  → Chatterbox TTS
  → audio playback

Entirely local except the Claude API call. Voice queries from Mission Control work hands-free.

📋 Page Groups

• Core: Dashboard, Feed, Briefing, System
• Work: Tasks, Calendar, Projects, Docs
• Business: Business, Invoices, Contracts, People
• Team: Agents, Team, Channels
• Analytics: Trading, Memory
• Life: Family, Teaching, Home Assistant

🎯 Why a dashboard?

Chat interfaces are great for conversation. They're terrible for seeing the state of your world.

• Live task status without asking
• Agent team overview
• Invoice + contract status
• Family calendar at a glance
• Voice queries from any page
• Home automation controls

What's integrated

• Bidirectional task sync via Graph API
• OAuth auto-refresh (no manual re-auth)
• Barack can create/update/complete tasks
• Mission Control Tasks page pulls live
• Planner ↔ n8n workflows for automation

What's connected

• Lights, climate, locks, automations
• Full bidirectional (read + control)
• Mission Control has a Home page
• Voice queries: "Is the garage door open?"
• Automations triggered by agent actions

Real automations

• Morning briefing → lights to 100% in kitchen
• "I'm leaving" → lock all doors, lower thermostat
• Kids' bedtime → Kids' room lights dim at 8 PM
• Late night work → office stays lit, rest of house dims

The problem it solves

Adam gets email from Agilent, U of I, clients, students, vendors, and mailing lists. Reading all of it is a full-time job.

The Inbox agent

• Polls Apple Mail every 30 min via cron
• Haiku classifies: urgent / action / FYI / noise
• Routes action items to Barack
• Sends daily digest to Signal at 9 AM
• Student emails → Sage gets context

Sample digest output

📄 Invoice Ninja

• Self-hosted at clients.yourdomain.com
• Sterling can create/track invoices
• Payment status in Mission Control
• Cloudflare tunnel for public access
• Client portal for easy payment

✍️ DocuSeal

• Self-hosted at contracts.yourdomain.com
• Contract generation + e-signature
• Matlock reviews before sending
• Stored locally (not DocuSign's cloud)

Local-first principles

• TTS inference runs on the M4, not cloud
• Memory files stay on local hardware
• Obsidian syncs via iCloud (not third-party)
• Sensitive docs: DocuSeal on local Proxmox
• Business data: Invoice Ninja on local Proxmox

Network security

• OPNsense firewall (not consumer router)
• AdGuard DNS-level blocking
• Tailscale VPN for remote access
• Cloudflare tunnel only for business tools
• n8n, Home Assistant: VPN-only

What goes to the cloud

• Claude API — reasoning/generation only
• Signal — end-to-end encrypted
• iCloud — Obsidian sync (encrypted)
• Not: voice audio, memory files, contracts, financial data

📁 What's in it

• README.md — full architecture overview
• QUICKSTART.md — 30-min setup (Mac + Windows)
• agents/ — all 16 agent templates
• prompts/ — Claude Code build prompts
• docs/system-prd.md — 🆕 Full PRD: build everything from scratch
• docs/ — architecture + integration guides
• examples/ — real automations to copy

🚀 The prompts are the secret

Each prompt file is something you can literally paste into Claude Code to build a component of your system:

• build-mission-control.md
• build-agent-team.md
• build-voice-system.md
• build-integrations.md

Installation

# Install OpenClaw
brew install openclaw

# Initialize workspace
mkdir ~/clawd && cd ~/clawd
openclaw init

# Start the gateway
openclaw gateway start

Your first 3 agents

# AGENT 1: Chief of Staff
name: Chief
model: claude-sonnet-4-5
system: "You are my Chief of Staff.
Route requests intelligently..."

# AGENT 2: Researcher  
name: Researcher
model: claude-sonnet-4-5
system: "You are a research analyst..."

# AGENT 3: Writer
name: Writer  
model: claude-opus-4-5
system: "You write clearly and..."

Build Mission Control

Paste prompts/build-mission-control.md into Claude Code:

claude code
> Read the OpenClaw docs at ...
> Build a Next.js dashboard with:
> - Agent status page
> - Task management
> - Memory viewer
> - Voice query support
> Dark theme, Tailwind CSS

3 iterations later: you have a real dashboard.

Add Chatterbox TTS

# Install Chatterbox
pip install chatterbox-tts

# Clone a voice (30s sample)
chatterbox clone \
  --input sample.wav \
  --name myvoice

# Start server
chatterbox serve --port 4126

# Configure OpenClaw
tts.provider = chatterbox
tts.baseUrl = http://localhost:4126

The prompt engineering loop

• 1. Describe the system in detail
• 2. Give Claude Code access to docs
• 3. Let it build, run, test
• 4. Review output, refine
• 5. Repeat until it works

Good Claude Code prompts

You are building a Next.js dashboard
for an AI agent system. The system
uses OpenClaw (docs at [url]).

Build a page called "Agents" that:
- Shows all configured agents
- Displays model, status, last active
- Has a "Send message" button per agent
- Uses the OpenClaw REST API

Tech: Next.js 15, TypeScript, Tailwind
Dark theme: bg #0a0a0f, accent #00d9ff
Run on localhost:3000

✅ Agent system prompts

• Define persona first — who are they?
• Define domain — what do they own?
• Define boundaries — what do they NOT do?
• Define escalation — when do they defer?
• Give examples of good/bad responses
• Include memory context injection point

❌ Common mistakes

• Vague personas ("be helpful")
• No boundary definition (agent sprawl)
• No escalation path (infinite loops)
• Too many agents too soon
• No memory system (Groundhog Day)
• Using Opus for everything ($$$)

What n8n does

• Visual workflow automation
• Connects APIs without code
• Webhooks in and out of OpenClaw
• Schedule complex multi-step jobs
• Error handling + retry logic
• 350+ built-in integrations

Example: Morning Briefing Workflow

CRON (6:45 AM daily)
  → GET M365 Calendar (today's events)
  → GET Planner (open tasks)
  → GET Weather API
  → POST OpenClaw/Barack (briefing)
  → Chatterbox TTS (obama voice)
  → Sonos HTTP API (play in kitchen)
  → Done ✅

Built in n8n visual editor in ~2 hours. Runs every morning without me touching it.

What OpenClaw provides

• 🛣️ Agent session management + routing
• ⏰ Cron scheduler (fires morning briefings)
• 📡 Multi-channel: Signal, iMessage, Email, Web
• 🔊 TTS integration + voice pipeline
• 🧠 Memory injection + context management
• 🤖 Sub-agent spawning + ACP harness

Peter Steinberger

Creator of OpenClaw

The nervous system that makes the whole AI infrastructure possible. None of this works without OpenClaw's gateway, session model, and channel multiplexing.

🤖 Personal AI Infrastructure

The framework this entire system is built on. Open source, MIT licensed, actively developed.

🌐 Human 3.0 Project

The destination — what AI augmentation is actually building toward.

🧵 Fabric — 243+ Patterns

Extract wisdom, summarize content, analyze papers, improve writing. Synced locally to ~/clawd/skills/fabric/

📬 Unsupervised Learning Newsletter

Weekly signal-to-noise on AI, security, and human flourishing. One of the best in the space.

Miessler's Formula

• Policy — goals & identity (humans set this)
• State — current reality (AI gathers this)
• SOPs — approved execution pipelines (AI owns these)
• Action — automation executes, humans steer

🌅 Morning Briefing — Our SOP

📐 Every Workflow Is a SOP

• Email arrives → Inbox triages → routes to agent → action or archive
• Student submits → Sage analyzes → rubric applied → feedback drafted → Riley edits
• Client inquiry → Sterling qualifies → Matlock checks risk → Maxwell drafts response
• System alert → Forge diagnoses → fixes if within SOP → escalates if not

🔮 What This Unlocks

• Define the SOP once → agents execute forever
• Humans update the SOP → agents adapt instantly
• "What's the SOP for X?" → conversation, not meeting
• Every workflow is measurable, improvable, auditable
• The CEO (you) finally sees the whole operation

Old thinking

We have Marketing. Let's add AI to it.
We have Security. Let's add AI to it.
We have HR. Let's add AI to it.

→ AI as a tool in the existing structure

The Inversion 🔄

You have a graph of operations. Some nodes are marketing. Some are security. Some are HR.

AI is the container. Industries are use cases inside it.

Software isn't bought anymore — it's benchmarked against the node it replaces.

🧬 Human 3.0 — Your Daemon

• Broadcast your full capabilities, not just a resume
• AI substrate matches you with work that fits
• Multiple income streams: retainer + project + gig
• Be compensated for being your full self
• "The best tutor in the world becomes world famous — and gets paid for it"

How It Works

• 216 grants from grants.gov + NSF indexed
• Faculty matching across 672 researchers
• Grant detail pages with team assembly

5-Factor Scoring

• Keyword match — 30%
• Department fit — 25%
• Methodology — 20%
• Funding range — 15%
• Collaboration potential — 10%

📸 Live demo:
Grant Intelligence dashboard

3 Voice Clones

• 🎤 Obama — strategy & briefings
• 🎤 Taylor Swift — personal & lifestyle
• 🎤 Olivia Benson — work & investigations

TTS Routing

• Mac mini M4 ↔ Gaming PC RTX GPU
• Auto-routes based on load & quality needs

Where It Shows Up

• Morning briefing on Sonos speakers (multi-voice DJ format)
• Live voice notifications on Mission Control
• Voice chat in channels
• Coming: near-real-time with Haiku 4.5

Finance Dashboard

• Net worth timeline, savings rate, monthly trend
• Spending by category — over/under budget in real time
• 3-month forecast with CSS-only charts (no external libs)

🧮 Frank Barker, CPA

• AI agent with a 20-year-trusted-advisor persona
• "Your food budget is 14% over. Want to talk about it?"
• Works with Vega on big picture cash flow + trading P&L

Why Agents Need Personas

• You interact differently with a CPA than a trading bot
• Persona = consistent decision-making framework
• Easier to calibrate trust: "Would Frank approve this?"
• Students: Who is your Frank for your domain?

Automated Scans

• Pre-market, midday, after-hours (3x daily)
• Prediction markets: Polymarket + Kalshi
• Options flow analysis + macro signals

Risk Controls

• Confidence calibration + weekly reflection
• Paper trading via Alpaca (prove before real $)
• Matlock agent as independent reviewer

Why This Matters for MSBA

• Time-series analysis at scale
• Confidence calibration (are your models right?)
• Systematic over emotional
• Build the system, then trust the system

Three Search Modes

• BM25 — fast keyword matching
• Vector — semantic similarity (embeddings)
• Hybrid + Rerank — best of both + LLM ranking

Collections

• Obsidian vault — 360 notes
• Workspace — 1,180 files
• Memory — 35 daily journals
• SOPs — 11 operational procedures

Fully Local

• Runs on Mac mini M4 with Metal GPU
• 300M embedding model + 600M reranker
• 4,283 vector chunks indexed
• MCP integration (any agent can search)

SOPs = Agent Instructions

• Morning briefing — 7:15 AM daily, multi-voice DJ format
• Trading scans — 3x daily with risk controls
• GitHub publish — mandatory security scan before every push
• Research due diligence — tiered depth (5/15/30 min)
• Trade execution — position sizing, stop-loss, Matlock review

Built-in Security

• Pre-push hook — scans for API keys, tokens, PII
• Catches large files (>50MB) automatically
• Git history scrubbed when secrets found
• Secrets backed up locally, never in git
• All credentials via process.env, not hardcoded

Step 1 — Install (30 min)

Follow QUICKSTART.md in the repo

• Install OpenClaw + Claude Code
• Configure your Anthropic API key
• Start the gateway → open localhost:18789
• Works on Mac, Windows (WSL2), Linux

Step 2 — Onboarding Interview (15 min)

Paste this into OpenClaw's web UI:

Read ONBOARDING.md and run me through
the onboarding interview. I want to
design my personal AI system.

Your agent interviews you → designs your system → produces a Claude Code build prompt → Claude Code builds your dashboard while you sleep.

What the interview covers

• Who you are + biggest time sinks
• Your real tools (calendar, email, chat)
• Your setup (laptop/server, budget)
• Which agents you actually need (3-8, not 16)
• Which dashboard pages make sense for you
• Your first automation

What you get out

• Personalized SOUL.md + USER.md
• Custom agent configs (your team, your names)
• Claude Code prompt → builds your dashboard
• First cron job ready to paste in
• Numbered next-steps list, specific to you

Where this is going

• AI infrastructure will be table stakes for knowledge workers
• The people who understand systems thinking win
• MSBA skills — data, analytics, architecture — are exactly what this requires
• The gap between "AI user" and "AI builder" is a career moat

Your advantage

• You understand data pipelines
• You can evaluate AI output critically
• You know when to trust a model (and when not to)
• You can design the system, not just use it