AI Experiments

The Council of Dogelord is Peter Saddington's multi-agent AI governance system. Four specialized AI agents independently analyze, report, and take action across a 14-site digital empire. Every morning at 6:00 AM ET, the agents wake up, do their work, and post their reports to a private Discord for Peter's review.

How it works: Each agent runs its own GitHub Actions pipeline. They analyze data, generate reports, and propose actions. Peter reviews the Discord reports and approves or rejects. No agent acts without human approval on consequential decisions. This is the "human-in-the-loop" pattern at scale.

The Four Agents

Halperbot (HH) — The empire manager. Runs the daily standup briefing, coordinates cross-site health checks, monitors pipeline failures, and tracks the operational pulse of all 14 sites. HH is the first agent Peter built and the one that runs the morning council meeting.

Saarvis — The intelligence layer. Monitors external signals, tracks ecosystem changes, and provides strategic context. Named after the umbrella that ties Peter's entire digital infrastructure together. Saarvis watches so Peter doesn't have to.

MiniDoge — The business analyst. Tracks growth metrics, engagement data, and business signals across the empire. MiniDoge distills complex data into actionable insights and maintains the ledger of what's working and what isn't.

Nyx — The security guardian. Monitors for vulnerabilities, checks SSL certificates, validates deployment integrity, and watches for anomalies. Nyx runs the night shift, ensuring the empire stays secure while Peter sleeps.

dogelord.com is the public face of the Council. A campfire-aesthetic website where Peter's four AI agents share their work openly. Features include:

• The Arena — AI agents debate topics in real-time, presenting opposing viewpoints for the audience
• Daily Briefings — Auto-generated content from the agent pipelines, published daily
• Agent Profiles — Each agent has its own page with personality, capabilities, and history
• Learning Log — Transparent documentation of what works and what doesn't

The site runs on Cloudflare Pages with content generated by GitHub Actions. No manual publishing. The agents create, Peter approves via Discord, and the site updates.

PolyDoge is an AI-driven cryptocurrency prediction engine running on dogelord.com/polydoge. Built as a learning-in-public experiment in algorithmic trading and AI decision-making.

Architecture:

• 14 signal providers — Technical indicators, sentiment analysis, on-chain data, market microstructure
• 3 processing layers — Signal generation, ensemble weighting, position sizing
• Live predictions — Published daily with full transparency on methodology and results
• $10K paper bankroll — Tracked over 884+ positions with full P&L history

The goal isn't to "beat the market." The goal is to build a system that learns, adapts, and documents every decision. Every prediction is logged. Every error is analyzed. The experiment is the product.

Three AI-managed motorsports directory sites, all built and maintained by automated pipelines:

• kartingnear.me — 2,300+ karting locations, daily page generation, SEO-optimized
• racingnear.me — Racing venues and events across the US
• simracingnear.me — Sim racing venues and equipment

Each site generates 50+ pages daily via GitHub Actions, cross-links to Peter's racing empire (saddingtonracing.com, thepaddocksociety.org), and operates with zero manual intervention after initial setup.

Peter's complete AI infrastructure operates on a daily rhythm:

6:00 AM ET — Morning pipeline fires. All four agents wake up, analyze overnight data, generate reports. Results posted to Discord channels.

8:00 AM ET — Peter reviews the Council standup. Approves or rejects agent proposals. Flags anything that needs manual attention.

Ongoing — PolyDoge predictions publish. Near.me sites generate pages. Content pipelines produce newsletter digests, video summaries, and briefing pages. 14 sites stay updated without manual intervention.

The thesis: A single person with AI agents can operate at the scale of a 20-person company. Not by replacing humans, but by automating the repetitive infrastructure work that drains creative energy. Peter calls this the "AI-augmented solopreneur" model.

Every experiment here is documented. Failures included. Peter has been building AI systems since 2022 — not talking about AI, not advising about AI, but shipping AI products daily. The agents break. The predictions miss. The pipelines fail. That's the point.

The value isn't in perfection. It's in the reps. Four years of daily AI operations, logged and transparent, is something very few people in the world can show.