Seventeen years turning manual operations into automated systems. I run enterprise endpoint, cloud, and security infrastructure, and I build LLM tooling into it to remove the manual, repetitive work that survived every earlier attempt to automate it.
I was the one who got paged.
Most people building with AI right now have never operated anything that couldn't afford to go down. I have. When a mortgage lender's endpoints, identity, or infrastructure broke, my answer was always the same: automate the failure point so it can't happen twice.
Read, judge, and decide. Those are in range now.
What changed is the class of problem I can reach. The processes that resisted scripting were the ones that needed someone to read, judge, and decide. I'm not reinventing myself as an AI researcher — I want to be the systems person who knows exactly where AI belongs in production infrastructure, and where it doesn't.
Find the manual, repetitive, error-prone process. Replace it with something reliable. That's been the job all along — AI is the first tool that reaches the steps scripting never could, and knowing which steps those are is the part you can't learn from a course.
Search over-fetches by cosine similarity, then re-ranks on a blend of similarity, literal word overlap, and title match — because pure vector search drifts on short queries. And an answer that arrives without a source marker is withheld rather than shown, so the system fails loudly instead of confabulating quietly.
A scheduled job runs at 2 AM, finds older notes that still match what I wrote that day, and writes questions into the next day's file.
Three shipping side projects. Two have nothing to do with AI — they're here because the layer underneath the model is the layer that decides whether anything you build on top of it stays up. The third is the ingestion pipeline feeding the retrieval system above.
Transfers a file between two Android phones over nothing but light — Luby Transform fountain codes rendered as QR frames at 30 fps, rebuilt by the receiver's camera. No network, no pairing, no back-channel: dropped frames become fewer frames rather than failures.
A Windows disk space analyzer that scanned 12 GB across roughly 50,000 files in 616 ms — a lock-free work stack drained by up to 64 threads, directory enumeration batched at ~600 entries per syscall, and value-type entries keeping the garbage collector out of the hot path.
The ingestion layer for the retrieval system above, packaged so anyone can run it. Deliberately not a summarizer — a summary is a worse copy of the source, and nothing ever links to it. It extracts claims: one idea per note, titled as an assertion, citing the page or timestamp it came from.
The graph standing behind this page is baked from the vault as of September 9, 2026 — the structure the time-lapse above showed forming, standing complete.
Campus helpdesk at Stevenson University — hardware, software, network, and account issues for a high-volume, diverse user base, plus the customer-service instincts a pure technical career rarely teaches.
Stevenson University · Maryland.
Owned the endpoint, cloud, and security posture of a financial services organization's device fleet end to end — a hybrid Intune / SCCM environment, Tanium at fleet scale, Azure VMs with backup and disaster recovery, a hardened security stack, and PowerShell automation across Microsoft 365.
Designed and built an internal AI-powered IT operations assistant — the first time the automation could handle the judgment step, not just the click.
Microsoft Azure Fundamentals coursework.
The infrastructure stack is a career deep. The AI stack is about two years deep, and listed honestly — what I've built with, not what I've read about.
Tanium — patching, compliance, scripting — plus its GraphQL API; Microsoft Intune; SCCM for imaging and software deployment.
Microsoft Azure — VMs, backup & recovery; VMware; Hyper-V.
CrowdStrike Falcon; Druva cloud backup; GlobalProtect VPN; Credential Guard, Windows Hello, MFA.
PowerShell; Python; REST & GraphQL API integration; Tanium API development; PRTG Network Monitor; Microsoft 365 & identity.
Retrieval-augmented generation; embeddings, vector stores, chunking; hybrid search & re-ranking; prompt and system-prompt engineering; grounding with enforced citations; local models — Ollama, ChromaDB.
LLM tool calling & tool schemas; Google Gemini AI Studio; Tanium / Zendesk tool integrations; Cloud Run; agentic coding harnesses; reliability & self-correction patterns.
I'm looking for a systems, infrastructure, or platform role where building with AI is part of the job rather than a side project — Senior Systems Administrator, Infrastructure or Platform Engineer, IT Automation, AIOps. The work I want is the long-standing architecture and efficiency problem everyone has quietly learned to live with, solved with AI tooling by someone who'll still be accountable for it at 2 AM.
