BMD Pat LLC
Patrick Hughes
Builder of a local AI sizing desk · Nashville, TN
I help people match the GPU they own to a model that will run. I measure the result on controlled compute, publish the evidence, and keep the useful tools small.
Claude Certified ArchitectAZ-305 · Azure Solutions Architect Expert
AI-102 · Azure AI Engineer Associate
Anthropic · Claude Certified Architect (Professional)
Start here
See what actually ran.
Read the benchmark rows and failed runs, or see how I stop an agent before it runs up a bill or gets stuck in a loop.
Who I am
I'm Patrick Hughes. I run BMD Pat LLC from Nashville and build local-first AI tools around a simple question: what will run on the hardware you already own?
I came up through IT infrastructure, went deep on Azure and AI engineering, and earned the certs to prove it (Azure Solutions Architect Expert and AI Engineer Associate). The side business started with a crypto mining rig in 2017 and has compounded ever since: open source tools, automated trading, angel bets, a 199-episode podcast, and now a local AI lab running on my own GPUs.
Scheduled agents handle repeatable operations. I make the decisions, own the public and money gates, and keep the system pointed at freedom through ownership. When I'm not doing either, I'm in a Dota match.
How BMD Pat runs itself
BMD Pat is human-led and queue-driven. Agents publish the blog, process research, maintain product surfaces, and surface failed runs. The morning brief arrives at 7:30 AM Central with what shipped, what failed, what needs a decision, and whether the machine is healthy.
I work on the business. They work in it. Products stay small, purchases stay async, and hiring is capital-gated.
07:30
- What shipped
- What failed
- What needs a decision
- Whether the machine is healthy
08:30
- Reviews queued drafts
- Approves or sends back with reasons
08:55
- Applies the QA notes
- Resubmits for review
09:30
- Publishes approved drafts only
- Cross-posts to Dev.to09:45
human gate
- Anything public
- Anything that spends money
- Anything that changes direction
Track record
AgentGuard
Open source · pip install agentguard47
A Python runtime guardrail SDK for AI agents: budget, loop, timeout, and rate limits before a run touches real work.
Entrefied podcast
199 episodes · 2016 to 2020
A long-running show on building businesses. Still pulling evergreen listens years after the last episode.
Local AI lab
Sizing desk · controlled compute
The desk turns owned hardware into a model-fit answer. The 5090 Reports preserve the measurements and failed runs.
Angel portfolio
Equity crowdfunding · values-aligned
A small VC-style book across health-tech, platforms, and safety. Most go to zero; a few do not.
Azure certified
AZ-305 Expert · AI-102 Associate
Solutions Architect Expert and AI Engineer Associate, verifiable on Microsoft Learn.
Building in public
Blog · GitHub · real numbers
Shipped tools, failed runs, and the systems that run the company, documented with real numbers or none.
Org Chart
BMD HODL Organizational Structure
22 registered operators. Human-led. Hiring is capital-gated.
Patrick Hughes
Owner / final judgment and public gates
No VP of Digital Transformation. No Center of Excellence for AI Enablement. No Change Advisory Board review pending.
operations
Morning Briefing
Chief of Staff
CMO
Chief Marketing Officer
Autotron
Head of Operations
Reflection
Head of Strategy
First-Paying-Customer
Head of Sales
finance
CFO
Chief Financial Officer
Stock Briefing
Equity Analyst
Crypto Briefing
Crypto Analyst
Autotrader
Portfolio Manager
growth
engineering
SDK Developer
Senior SDK Engineer
Dashboard Developer
Full-Stack Engineer
Marketing & Growth
Docs & Growth Lead
Project Manager
PM
infrastructure
Security Analyst
P0 Threat Detection
NightShift Supervisor
After-Hours Monitor
Email Router
Inbound Triage
Cleanup
Repo Hygiene
Monitoring
System Health
Cron Supervisor
Scheduler Oversight
FullAutoResearch
Research Director
BMDHodl Agent
Asset Tracker
People operations — Hiring stays closed until the capital gates pass.
Current rung: 0 human hires. The operating system remains automated.
The rig
Why I'm doing this
Freedom is the destination. Ownership is the thesis. The current Builder is the local AI sizing desk at /desk. The 5090 Reports are its measurement engine, while AgentGuard stays in maintenance mode and the Workflow Audit remains a legacy async door. The first operating floor is $200 a month, but the goal is owned systems that buy back time.
Ship things people actually use. Repeat. Keep the surface small enough that a human owner can understand the important decisions while agents carry the routine work.
“I build systems that give people tools, capital, and freedom at scale.”
What I'm NOT
Not a consultant. Not an agency. Not taking calls. I'm not going to write you a proposal. I'm not going to align on scope over a 30-minute discovery session.
I ship tools. You buy them, or you don't. That's the whole relationship, and it's the one I wanted.
Skills & Stack
Background
I came up through IT infrastructure and systems engineering: cloud migrations, network operations, and regulated, high-stakes environments. Then I went deep on Azure and AI engineering. Fifteen years of turning messy systems into machines that keep running without me.
The certs are the paper trail (Azure Solutions Architect Expert and AI Engineer Associate, both verifiable on Microsoft Learn). The tools on this site are the current work: open models on owned hardware, measured honestly, packaged into products a stranger can use in under a minute.