I came up through oil and gas, from field hand to president of an engineering services company. Since then I've put 10,000+ hands-on hours into AI models, building with them and documenting where they misbehave.

That work runs on two tracks:

• Scaling Success: my practice. I build small, portable tools with AI, fast, alongside the team that will own them, so they can do the next one without me.
• Field notes: reproducible write-ups of how LLMs and AI dev tools behave at the edges. Prompt injection, guardrails that block actions but not outcomes, agents that route around their own restrictions.

I also run The Warden, a prompt injection CTF with stacked defenses on live models. It's free to play, and I run guided sessions and workshops on it, for teams, groups, and organizations.

Programmable Execution & Processing for Project Engineering & Robotics

Pepper is a durable, and proactive personal AI assistant: Presented as an animated 3D robot on my desktop with one long-running core behind it that owns memory, routing, and every model call. It started as a knowledge-base idea and turned into an AI operations layer.

• Routes each request to the best-fit model across Anthropic, OpenAI, Cursor, and OpenRouter, with automatic fallback when a plan hits its usage cap and queued turns that resend once the cap lifts.
• Remembers across projects, models, and devices. One thread per project, SQLite memory that gets redacted before it's stored, the active project picked up from the focused window, and Claude Code transcripts imported.
• Lives on the desktop as an animated 3D avatar on a Hyprland layer-shell overlay. It shows what Pepper is doing, looks, follows, raises notices, briefs, nudges, and follow-ups, and lets me answer approvals in place.
• Runs always-on. The core sits on a dedicated host, and runners on each workstation execute turns locally. Remote access goes over Tailscale, limited to trusted devices.
• Works unattended, but never ships on its own. Watches on GitHub and Railway open tickets. Those run sandboxed in clones, ask the router before each tool call, and stay inside subscription limits. Nothing is pushed without my approval, and a ticket stays open until CI passes.

Written in Go, with 37 ADRs so far. The repo is private while it's under active development. Next up: voice, vision, and a phone client. Further out, the plan is a Raspberry Pi 5 robot body.

I currently have a pretty short list of public work, since a lot of my work is around research, cyber security, and pushing AI limits. A few of them are listed here. I am more open to share individually, so feel free to reach out or check out my research blog

• 📧 brian@scalingsuccess.io
• 🗓️ Want to work together? Book a call
• 💬 GitHub issues on any public repo work too

life_stats = {
    "background": "oil & gas operations",
    "hands_on_ai_hours": "12,000+",
    "primary_languages": ["Go", "TypeScript", "Python"],
    "os": "Arch Linux",
    "wm": "Hyprland",
    "ai_sidekick": "P.E.P.P.E.R.",
    "coffee_dependency": True,
}

󰣇 definitelynotnsa in ~/git_projects/Curious-Keeper on  main [!]
❯ git commit -m "trust me" && git push --force-with-lease  # growth