What I actually do

Running a security program is a leadership job before it's a technical one. I've built and led 24/7 global security operations teams, carried security through the Mylan-Viatris merger without disrupting operations, and now own the governance layer that decides what AI/ML work gets built, reviewed, and shipped across Viatris. Most of my job is translating technical risk into a decision an executive can act on.

I design AI systems to be secure from the first line of the architecture diagram, not patched after something breaks. In practice that means constrained outputs, confidence thresholds, and hallucination checks built into the pipeline itself, not bolted on afterward. At Viatris, I lead security architecture across the company's AI/ML initiatives; every tool I publish on this site is held to the same bar.

Most of my hands-on work is fine-tuning small, open models for one specific job instead of reaching for the biggest model available, then publishing the weights instead of just the results. I orchestrate multi-agent pipelines with LangGraph and smolagents, and I've written publicly about why a smaller, cheaper model usually beats a large one on a well-defined task.

Before any of this was AI, it was data analysis. I hold a master's in business data analytics from WVU, and I still build my own dashboards rather than handing that off, a habit that goes back to my first analyst role doing geopolitical risk work with R and Tableau.

I ran a 24/7 global security operations center for years, and OSINT and geopolitical risk analysis is where my career actually started, well before any of it was AI. The tools I build now point at that same problem: turning scattered intelligence into a decision someone can actually act on.