I learned to build systems first as mechanisms and music: preserve the object, respect the constraint, make the interface playable, then test against reality.
At sixteen I started designing physical tools for medical-device manufacturing at Applied Medical: pneumatic robotic tools, manipulators, fixtures, EOAT, CAD, prototypes, qualification, and shop-floor debugging. Complexity had to compress into reliability because the work sat next to production.
Before that, FIRST Robotics taught design under time pressure. FRC 3476 "Code Orange" won a World Championship, IRI Championship, and Industrial Design Award while I co-led mechanical design. The lesson that stayed: immature design tries to control too much, then compromises the parts that really do have to be controlled.
Through those years I played competitive piano, taught, improvised, and performed. Music gave me the other half of the method: structure first, expression inside structure, revision by attentive listening.
I eventually accepted that modeling intuition, not merely using it, was the problem I kept circling. UCSD became the formal phase: Cognitive Science with a Machine Learning & Neural Computation specialization, reinforcement learning with Mattar, neural data science with Voytek, and design-research work with Hyundai and Ford through The Design Lab.
At RAM Labs I worked across tactical cybersecurity, vulnerability repair, distributed edge inference, and proposal-driven research. I designed graph and transformer-family architectures, led a $250k SBIR as PI, co-invented a deep-learning bug-fixing patent, built pySABRE for blockchain consensus simulation, and represented the company at Google for DARPA FACT.
At a Drone Services Startup and through consulting, I shipped production systems end-to-end under aggressive cost and time constraints: architecture, implementation, deployment, and operations. The useful pattern is boring in the best way: stabilize the representation, expose uncertainty, and leave the human operator with a handle.
The through-line across thirteen years is not a genre. It is a way of decomposing ambiguous work into stable objects, typed relations, hard validators, and learned residuals. Frisbee intake. Piano motive. Pneumatic fixture. T-UEBA pipeline. CAD object graph. Same instinct under different consequences.
The strange parts should earn their place by making the practical parts clearer.