I got into AI/ML the long way around. I started in chemical engineering at the University of Washington, drawn to the parts of the curriculum that looked the most like applied math: process control, transport phenomena, numerical methods. After my BS, I joined a bioengineering research group at UW where I was the person who could read a differential equation and also write the Python that solved it. That turned into seven years of research engineering on point-of-care diagnostics, first the software and statistics, then the machine learning side of it.
Along the way I went back for an MS in Applied Mathematics. The research had pushed me into territory my BS hadn’t covered: convex optimization, numerical linear algebra, statistics. I wanted the formal foundation. That’s also when I started building the things I’d been doing in MATLAB in PyTorch instead. CNNs for diagnostic test image regression, ensemble models combining biometrics with binary clinical signals, real-time dashboards for experimental data.
Today I work on generative AI infrastructure at Meta: evaluation platforms for multimodal models, ML inference services running on H100 clusters, and the data pipelines that feed both. The throughline from chemical engineering to GenAI infra is shorter than it sounds: in both, the system you’re studying is too big to hold in your head all at once, you instrument it carefully, and you spend a lot of your time deciding which signals are real.
On the side, I build and ship products. The Axiomatiq ratings platform powers two live consumer sites: Tastemongers (cheese) and Edgemongers (chef’s knives). Both started as places to test whether agentic LLM workflows could produce useful structured data from messy sources. Both turned into real businesses with real customers.
You can find me at dannylleon@gmail.com or on GitHub and LinkedIn.