interrogate me

StevenHoggatt

Enterprise architect with 18+ years across DoD and Microsoft estates, currently building agentic AI infrastructure that does real work in a homelab that mirrors enterprise constraints.

MCP servers
Models loaded
Vectors
Last update
Token Burn
329.7M exact

background

The arc

While I was attending Angelo State University for a Bachelor of Management Information Systems (BMIS), I had two jobs through college. One was at a local computer repair shop building computers and fixing the various issues that came in. The other was at a local Internet service provider (ISP) doing phone support. This was before scripts and guides were available, when places like modemhelp.net were the reference and "line noise" was something you diagnosed by running the modem's diagnostic AT commands. US Robotics modems had great ones. I helped people who couldn't read get connected, and explained to more than one customer why their neighbor was disconnecting their connection: a Plain Old Telephone Service (POTS) line on a shared pair.

After graduation in 2003 I moved back to San Antonio and got a job at Altex Electronics. I figured every IT shop in the city eventually goes to Altex, which turned out to be true. That's where I met the people from McCombs Enterprises. I went to work for McCombs for about three years on a four-man team supporting around 15 locations, a mix of domain-joined and workgroup systems. From there I went to a company with an Air Force contract and worked my way up from helpdesk to system engineer. That led to a year in Afghanistan supporting the 4-401st logistics division (ask AI how going to Afghanistan came to be!).

Came home and went to work for a managed service provider (MSP), where I started my career in Office 365 (O365) and Azure migrations. That's also where I picked up the school district, migrating them from Novell/GroupWise to Active Directory (AD), Active Directory Federation Services (ADFS), and Azure/O365, and standing up System Center Configuration Manager (SCCM), now Microsoft Endpoint Configuration Manager (MECM). Then eventually migrating them off ADFS as Entra ID matured enough and the new auth method provided better security. The district became a long-running parallel track for me; I've stayed with them since 2015 because it gives me architectural latitude Department of Defense (DoD) work doesn't, and lets me try things end-to-end I'd only get to design in a federal environment.

From the MSP I got hired back into a DoD entity for an SCCM migration project, part of the broader DoD consolidation effort moving fractured service under one agency. Once that was done I rolled into sustainment maintaining the endpoint platform, then transitioned to another company as a solution architect. Initially I was helping teams build Splunk dashboards; from there I moved to the teams building out the cloud environment and the governance around it.

Most recently I transitioned to a new team focused on AI. I'm currently architecting an AI implementation for a documentation pipeline that ingests chats, documents, and meeting transcripts and has AI build out the documentation and organize the meetings and chats with Mermaid diagrams. That's the spine of what I do now, and it's continuous with everything that came before. Identity, endpoint, governance, automation, just with agentic AI as the new orchestration surface.

When AI stopped being a hobby

I've been working with AI seriously since early 2025, starting with ChatGPT and building my own custom GPTs. Retrieval-augmented generation (RAG)-style setups for quick reference to things like my tractor's owner's manual. In November 2025 I decided to complete the lab environment and go "all-in." I loaded Claude Code and Codex in the command-line interface (CLI) and have been building in my homelab ever since: a personal RAG memory layer over my home telemetry, a Model Context Protocol (MCP) server that protects context windows for large-output tool calls, an agentic docs pipeline pattern, a NotebookLM-style podcast generator that can de-vendor source material and fill author gaps with retrieved context, and the rest of what's in the next section.