Systems Architect. 12 years in hosting, 24 in IT. Currently building tools that intersect with AI, security, and civic tech.
I've worked in IT for over 20 years. I am a Systems Architect at a web hosting company. I manage around 1,100 VMs across the brand (VMWare), plus overflow support for other brands in the parent company's portfolio. In my home lab, I use Docker. Currently sitting at 6 containers, 5 of which run PostgreSQL databases for different projects. I use local LLMs that run inference on my scaffolded setup for my AI agents. I started at the bottom floor as a T1 and have worked my way up, learning everything I could as I went. I am comfortable designing and troubleshooting at any position in the stack.
In my homelab, I have built a team of 5 AI agents that work together. Each with their own role and memory. Orchestration, code, backend design, frontend, and documentation. These each have directives pointing at my preferred style when it comes to CSS or how to structure code/documentation with references on what is a reliable "source" of information on their specialty. Additionally, I've made a process with its own set of tools to play red agent against my ideas. This takes a look at the project as a whole to poke holes and let me know where it needs hardening. Allowing me to hone in and review these more quickly and speeding up the iteration process.
My take or use of AI is a bit different than how it is commonly used or portrayed as used anyways. My specialty is making deterministic tools for AI to use. Since LLMs drift, hallucinate, or make up data or sources, creating deterministic tools that provide a calculated output greatly reduces the chances of that seeping into finished reports or output. Since AI can essentially become a digital Doc Ock, it has at least 8 hands to hold these tools. So every project I work on, I make small documented tools that give a deterministic output that the AI agents then use in tandem with each other. The toolbox continually grows, further constraining the LLM, reducing drift and at the same time lowering token usage. Pair this with a human review process (documented checklist and process) and then finally a red agent evaluation of my project and you end up with a tight end report that can then be used as reference when writing a report. I like to stay away from having AI generate the end content. It can do it, but not very well and with lots of holes. Having it part of the process is a huge boon though.
I'm active in civic tech. I built a political transparency database for Oklahoma, tracking state legislators' connections to businesses, donors, and lobbyists. I've also built another for Tulsa elected officials, this one being person-centric and focusing on conflict of interest. I've tried to keep the design principle simple in theory, but it gives lots to work with. Every fact has a source URL, a fetch date, and a confidence level that are recorded to a database. The confidence level uses several sources for this, including Google Scholar's h-index.
I'm pretty straightforward on where my values stand. I believe that infrastructure should serve the people and not extract. I'm interested in AI infrastructure, security engineering, DevOps, or civic tech. I'm based in Tulsa, OK.