Patrick Hughes

Manager of Software Engineering & AI @ OneOncology

Building AI-operated systems out of Nashville, TN

---

---

Day job: engineering + AI at a healthcare company.

After hours: running BMD HODL -- a one-person AI-operated holding company. 12 agents. One human. Compound across decades.

---

Open Source

AgentGuard   pip install agentguard47

Your agent just burned $200. AgentGuard stopped it at $5.

Runtime cost guardrails for AI agents. Budget caps, loop detection, kill switches. Drop it in front of any LLM call. Zero dependencies. MIT.

from agentguard47 import AgentGuard

guard = AgentGuard(budget=5.00)

with guard:
    response = client.chat(...)  # enforces budget at runtime

Works with OpenAI, Anthropic, LangChain, or anything that makes LLM calls.

---

Local AI Rig

RTX 3070 + RTX 5070 Ti + RTX 5090
llama.cpp running Llama 3.1 8B

Because sending every token to a cloud API gets expensive fast.

---

Writing

I write about AI agents, local LLMs, and building things solo at bmdpat.com.

• Prompt injection and AI agents: what actually breaks
• Local LLM inference on consumer GPUs in production
• AI agent cost control with AgentGuard

---