The full story.

I studied mechanical engineering at Cornell and stayed for a master's in aerospace, finishing the B.S. in May 2025 and the M.Eng. that December. Along the way I was given the Sibley School's Outstanding Senior Award. I now work as a sales engineer at Embraer Executive Jets in Melbourne, Florida, doing aircraft performance analysis and building the internal Python tools that make that analysis usable by people who are not engineers.

The thread through most of my work is the gap between what a model predicts and what an instrument reads. The first time that became concrete was the summer I spent in Cornell's Wind Tunnel Studio. Four tunnels had aged into producing invalid data, and I owned the piece of the rebuild that replaced a fixed pitot-static probe with a hot-wire traverse — designing it, machining the mounts myself, wiring the acquisition chain, and then surveying 154 points across the test section to prove the flow was uniform. That project taught me that a measurement is only as trustworthy as the hardware holding the sensor still.

The same summer produced the other half of the lesson. The team ran a turbine campaign through the rebuilt tunnel and measured peak power roughly 22% below prediction. I wrote a blade load and deflection model from scratch to test whether blade flex explained it — blade element momentum theory closed as a constrained minimisation, section properties from a Green's-theorem boundary integral, a piecewise Euler–Bernoulli beam solver. The answer was no: tip deflection came out at about 1.9% of span, far too small to account for the shortfall. A model that rules out your favourite hypothesis is doing its job.

My M.Eng. work with Professor Jane Wang pushed further in the analytical direction. I derived the nonlinear equations of motion for a flapping-wing flyer from Newton's second law and blade-element aerodynamics, built the simulation in Python, found a hovering-capable set of wing kinematics by parameter study, and then trained a PPO agent to search for them on its own. The agent improved substantially but did not reach sustained hover — and being able to say exactly why, and to point at the reward shaping rather than the physics, is the part of that project I am most confident about.

I like problems where the answer is checkable. I like being able to say what a number is worth and what it is not worth. And I like building the thing that produces the number in the first place.